| Line |
Branch |
Exec |
Source |
| 1 |
|
|
#include "ps/rdma/control_plane.h" |
| 2 |
|
|
|
| 3 |
|
|
#include <grpcpp/grpcpp.h> |
| 4 |
|
|
|
| 5 |
|
|
#include <chrono> |
| 6 |
|
|
#include <cstdint> |
| 7 |
|
|
#include <cstring> |
| 8 |
|
|
#include <memory> |
| 9 |
|
|
#include <mutex> |
| 10 |
|
|
#include <stdexcept> |
| 11 |
|
|
#include <string> |
| 12 |
|
|
#include <utility> |
| 13 |
|
|
|
| 14 |
|
|
#include "base/log.h" |
| 15 |
|
|
#include "rdma_control_plane.grpc.pb.h" |
| 16 |
|
|
|
| 17 |
|
|
namespace petps { |
| 18 |
|
|
|
| 19 |
|
|
namespace { |
| 20 |
|
|
|
| 21 |
|
|
using recstoreps::rdma::GetMetaRequest; |
| 22 |
|
|
using recstoreps::rdma::GetMetaResponse; |
| 23 |
|
|
using recstoreps::rdma::GetServerHealthRequest; |
| 24 |
|
|
using recstoreps::rdma::GetServerHealthResponse; |
| 25 |
|
|
using recstoreps::rdma::ProbeRequest; |
| 26 |
|
|
using recstoreps::rdma::ProbeResponse; |
| 27 |
|
|
using recstoreps::rdma::PublishMetaRequest; |
| 28 |
|
|
using recstoreps::rdma::PublishMetaResponse; |
| 29 |
|
|
using recstoreps::rdma::PublishServerDrainingRequest; |
| 30 |
|
|
using recstoreps::rdma::PublishServerDrainingResponse; |
| 31 |
|
|
using recstoreps::rdma::PublishServerReadyRequest; |
| 32 |
|
|
using recstoreps::rdma::PublishServerReadyResponse; |
| 33 |
|
|
using recstoreps::rdma::RawVerbsMetadata; |
| 34 |
|
|
using recstoreps::rdma::RdmaControlPlane; |
| 35 |
|
|
using recstoreps::rdma::WaitServerReadyRequest; |
| 36 |
|
|
using recstoreps::rdma::WaitServerReadyResponse; |
| 37 |
|
|
using recstoreps::rdma::WaitServerRequest; |
| 38 |
|
|
using recstoreps::rdma::WaitServerResponse; |
| 39 |
|
|
|
| 40 |
|
60 |
std::string EndpointString(const RdmaControlPlaneEndpoint& endpoint) { |
| 41 |
2/4
✓ Branch 2 taken 60 times.
✗ Branch 3 not taken.
✓ Branch 5 taken 60 times.
✗ Branch 6 not taken.
|
120 |
return endpoint.host + ":" + std::to_string(endpoint.port); |
| 42 |
|
|
} |
| 43 |
|
|
|
| 44 |
|
92 |
std::chrono::system_clock::time_point DeadlineFromNow(int timeout_ms) { |
| 45 |
|
92 |
return std::chrono::system_clock::now() + |
| 46 |
1/2
✓ Branch 2 taken 92 times.
✗ Branch 3 not taken.
|
276 |
std::chrono::milliseconds(timeout_ms); |
| 47 |
|
|
} |
| 48 |
|
|
|
| 49 |
|
26 |
std::string GrpcStatusText(const grpc::Status& status) { |
| 50 |
1/2
✗ Branch 3 not taken.
✓ Branch 4 taken 26 times.
|
26 |
if (status.error_message().empty()) { |
| 51 |
|
✗ |
return status.error_code() == grpc::StatusCode::OK |
| 52 |
|
|
? std::string("OK") |
| 53 |
0/10
✗ Branch 0 not taken.
✗ Branch 1 not taken.
✗ Branch 4 not taken.
✗ Branch 5 not taken.
✗ Branch 8 not taken.
✗ Branch 9 not taken.
✗ Branch 10 not taken.
✗ Branch 11 not taken.
✗ Branch 13 not taken.
✗ Branch 14 not taken.
|
✗ |
: std::to_string(status.error_code()); |
| 54 |
|
|
} |
| 55 |
|
26 |
return status.error_message(); |
| 56 |
|
|
} |
| 57 |
|
|
|
| 58 |
|
78 |
void ThrowIfNotOk(const grpc::Status& status, const std::string& operation) { |
| 59 |
2/2
✓ Branch 1 taken 52 times.
✓ Branch 2 taken 26 times.
|
78 |
if (status.ok()) { |
| 60 |
|
52 |
return; |
| 61 |
|
|
} |
| 62 |
1/2
✓ Branch 1 taken 26 times.
✗ Branch 2 not taken.
|
26 |
throw std::runtime_error( |
| 63 |
4/8
✓ Branch 2 taken 26 times.
✗ Branch 3 not taken.
✓ Branch 5 taken 26 times.
✗ Branch 6 not taken.
✓ Branch 8 taken 26 times.
✗ Branch 9 not taken.
✓ Branch 11 taken 26 times.
✗ Branch 12 not taken.
|
52 |
"control-plane " + operation + " failed: " + GrpcStatusText(status)); |
| 64 |
|
|
} |
| 65 |
|
|
|
| 66 |
|
6 |
std::string EncodeMetaBytes(const RawVerbsNodeMeta& meta) { |
| 67 |
|
6 |
return EncodeRawVerbsNodeMeta(meta); |
| 68 |
|
|
} |
| 69 |
|
|
|
| 70 |
|
4 |
RawVerbsNodeMeta DecodeMetaBytes(const std::string& payload) { |
| 71 |
|
4 |
return DecodeRawVerbsNodeMeta(payload); |
| 72 |
|
|
} |
| 73 |
|
|
|
| 74 |
|
2 |
grpc::Status MakeDeadlineExceeded(const std::string& message) { |
| 75 |
|
2 |
return grpc::Status(grpc::StatusCode::DEADLINE_EXCEEDED, message); |
| 76 |
|
|
} |
| 77 |
|
|
|
| 78 |
|
6 |
grpc::Status MakeUnavailable(const std::string& message) { |
| 79 |
|
6 |
return grpc::Status(grpc::StatusCode::UNAVAILABLE, message); |
| 80 |
|
|
} |
| 81 |
|
|
|
| 82 |
|
|
} // namespace |
| 83 |
|
|
|
| 84 |
|
|
// One channel/stub per client. A shard handshakes and heartbeats, so opening a |
| 85 |
|
|
// connection per RPC would be wasteful; gRPC channels are thread-safe and serve |
| 86 |
|
|
// concurrent RPCs, so every RPC of one client shares this cache. Creating a |
| 87 |
|
|
// channel does not connect, so an unused client costs nothing. |
| 88 |
|
|
struct RdmaControlPlaneChannelCache { |
| 89 |
|
34 |
explicit RdmaControlPlaneChannelCache( |
| 90 |
|
|
const RdmaControlPlaneEndpoint& endpoint) |
| 91 |
2/4
✓ Branch 2 taken 34 times.
✗ Branch 3 not taken.
✓ Branch 6 taken 34 times.
✗ Branch 7 not taken.
|
68 |
: stub(recstoreps::rdma::RdmaControlPlane::NewStub(grpc::CreateChannel( |
| 92 |
2/4
✓ Branch 1 taken 34 times.
✗ Branch 2 not taken.
✓ Branch 4 taken 34 times.
✗ Branch 5 not taken.
|
68 |
EndpointString(endpoint), grpc::InsecureChannelCredentials()))) {} |
| 93 |
|
|
|
| 94 |
|
|
// The generated stub keeps the channel alive. |
| 95 |
|
|
std::unique_ptr<recstoreps::rdma::RdmaControlPlane::Stub> stub; |
| 96 |
|
|
}; |
| 97 |
|
|
|
| 98 |
|
10 |
std::string EncodeRawVerbsNodeMeta(const RawVerbsNodeMeta& meta) { |
| 99 |
1/2
✓ Branch 1 taken 10 times.
✗ Branch 2 not taken.
|
10 |
RawVerbsMetadata wire; |
| 100 |
1/2
✓ Branch 1 taken 10 times.
✗ Branch 2 not taken.
|
10 |
wire.set_magic(kRawVerbsMetadataMagic); |
| 101 |
1/2
✓ Branch 1 taken 10 times.
✗ Branch 2 not taken.
|
10 |
wire.set_protocol_version(meta.protocol_version); |
| 102 |
1/2
✓ Branch 1 taken 10 times.
✗ Branch 2 not taken.
|
10 |
wire.set_deployment_id(meta.deployment_id); |
| 103 |
1/2
✓ Branch 1 taken 10 times.
✗ Branch 2 not taken.
|
10 |
wire.set_deployment_epoch(meta.deployment_epoch); |
| 104 |
1/2
✓ Branch 1 taken 10 times.
✗ Branch 2 not taken.
|
10 |
wire.set_configuration_digest(meta.configuration_digest); |
| 105 |
1/2
✓ Branch 1 taken 10 times.
✗ Branch 2 not taken.
|
10 |
wire.set_fabric_digest(meta.fabric_digest); |
| 106 |
1/2
✓ Branch 1 taken 10 times.
✗ Branch 2 not taken.
|
10 |
wire.set_node_id(meta.node_id); |
| 107 |
1/2
✓ Branch 1 taken 10 times.
✗ Branch 2 not taken.
|
10 |
wire.set_logical_id(meta.logical_id); |
| 108 |
1/2
✓ Branch 1 taken 10 times.
✗ Branch 2 not taken.
|
10 |
wire.set_node_role(meta.node_role); |
| 109 |
1/2
✓ Branch 1 taken 10 times.
✗ Branch 2 not taken.
|
10 |
wire.set_lid(meta.lid); |
| 110 |
1/2
✓ Branch 1 taken 10 times.
✗ Branch 2 not taken.
|
10 |
wire.set_qpn(meta.qpn); |
| 111 |
1/2
✓ Branch 1 taken 10 times.
✗ Branch 2 not taken.
|
10 |
wire.set_psn(meta.psn); |
| 112 |
1/2
✓ Branch 1 taken 10 times.
✗ Branch 2 not taken.
|
10 |
wire.set_rkey(meta.rkey); |
| 113 |
1/2
✓ Branch 1 taken 10 times.
✗ Branch 2 not taken.
|
10 |
wire.set_base_addr(meta.base_addr); |
| 114 |
1/2
✓ Branch 1 taken 10 times.
✗ Branch 2 not taken.
|
10 |
wire.set_gid(meta.gid, sizeof(meta.gid)); |
| 115 |
1/2
✓ Branch 1 taken 10 times.
✗ Branch 2 not taken.
|
10 |
wire.set_port_num(meta.port_num); |
| 116 |
1/2
✓ Branch 1 taken 10 times.
✗ Branch 2 not taken.
|
10 |
wire.set_link_layer(meta.link_layer); |
| 117 |
1/2
✓ Branch 1 taken 10 times.
✗ Branch 2 not taken.
|
10 |
wire.set_gid_index(meta.gid_index); |
| 118 |
1/2
✓ Branch 1 taken 10 times.
✗ Branch 2 not taken.
|
10 |
wire.set_active_mtu(meta.active_mtu); |
| 119 |
1/2
✓ Branch 1 taken 10 times.
✗ Branch 2 not taken.
|
10 |
wire.set_fabric_mode(meta.fabric_mode); |
| 120 |
1/2
✓ Branch 1 taken 10 times.
✗ Branch 2 not taken.
|
10 |
wire.set_lane_generation(meta.lane_generation); |
| 121 |
|
10 |
std::string payload; |
| 122 |
2/4
✓ Branch 1 taken 10 times.
✗ Branch 2 not taken.
✗ Branch 3 not taken.
✓ Branch 4 taken 10 times.
|
10 |
if (!wire.SerializeToString(&payload)) { |
| 123 |
0/2
✗ Branch 2 not taken.
✗ Branch 3 not taken.
|
✗ |
throw std::runtime_error("failed to serialize RawVerbsNodeMeta"); |
| 124 |
|
|
} |
| 125 |
|
20 |
return payload; |
| 126 |
|
10 |
} |
| 127 |
|
|
|
| 128 |
|
12 |
RawVerbsNodeMeta DecodeRawVerbsNodeMeta(const std::string& payload) { |
| 129 |
1/2
✓ Branch 1 taken 12 times.
✗ Branch 2 not taken.
|
12 |
RawVerbsMetadata wire; |
| 130 |
2/2
✓ Branch 2 taken 10 times.
✓ Branch 3 taken 2 times.
|
24 |
if (payload.empty() || payload.size() > kMaxRawVerbsMetadataBytes || |
| 131 |
3/4
✓ Branch 1 taken 10 times.
✗ Branch 2 not taken.
✓ Branch 3 taken 8 times.
✓ Branch 4 taken 2 times.
|
10 |
!wire.ParseFromString(payload) || |
| 132 |
2/4
✓ Branch 1 taken 8 times.
✗ Branch 2 not taken.
✓ Branch 3 taken 8 times.
✗ Branch 4 not taken.
|
8 |
wire.magic() != kRawVerbsMetadataMagic || |
| 133 |
4/6
✓ Branch 1 taken 8 times.
✗ Branch 2 not taken.
✓ Branch 3 taken 6 times.
✓ Branch 4 taken 2 times.
✓ Branch 5 taken 6 times.
✗ Branch 6 not taken.
|
14 |
wire.protocol_version() != recstore::kRdmaDeploymentProtocolVersion || |
| 134 |
4/8
✓ Branch 1 taken 6 times.
✗ Branch 2 not taken.
✓ Branch 5 taken 6 times.
✗ Branch 6 not taken.
✓ Branch 7 taken 6 times.
✗ Branch 8 not taken.
✓ Branch 9 taken 6 times.
✗ Branch 10 not taken.
|
18 |
wire.deployment_id().empty() || wire.deployment_epoch() == 0 || |
| 135 |
3/6
✓ Branch 1 taken 6 times.
✗ Branch 2 not taken.
✓ Branch 5 taken 6 times.
✗ Branch 6 not taken.
✓ Branch 8 taken 6 times.
✗ Branch 9 not taken.
|
12 |
wire.configuration_digest().empty() || wire.fabric_digest().empty() || |
| 136 |
4/8
✓ Branch 1 taken 6 times.
✗ Branch 2 not taken.
✓ Branch 3 taken 6 times.
✗ Branch 4 not taken.
✓ Branch 6 taken 6 times.
✗ Branch 7 not taken.
✓ Branch 8 taken 6 times.
✗ Branch 9 not taken.
|
6 |
wire.node_id() < 0 || wire.node_id() > UINT16_MAX || |
| 137 |
2/4
✓ Branch 1 taken 6 times.
✗ Branch 2 not taken.
✓ Branch 3 taken 6 times.
✗ Branch 4 not taken.
|
6 |
wire.node_role() > |
| 138 |
|
6 |
static_cast<std::uint32_t>(recstore::RdmaNodeRole::kClient) || |
| 139 |
4/8
✓ Branch 1 taken 6 times.
✗ Branch 2 not taken.
✓ Branch 3 taken 6 times.
✗ Branch 4 not taken.
✓ Branch 6 taken 6 times.
✗ Branch 7 not taken.
✓ Branch 9 taken 6 times.
✗ Branch 10 not taken.
|
6 |
wire.lid() > UINT16_MAX || wire.gid().size() != 16 || |
| 140 |
4/8
✓ Branch 1 taken 6 times.
✗ Branch 2 not taken.
✓ Branch 3 taken 6 times.
✗ Branch 4 not taken.
✓ Branch 6 taken 6 times.
✗ Branch 7 not taken.
✓ Branch 8 taken 6 times.
✗ Branch 9 not taken.
|
6 |
wire.port_num() == 0 || wire.port_num() > UINT8_MAX || |
| 141 |
4/8
✓ Branch 1 taken 6 times.
✗ Branch 2 not taken.
✓ Branch 3 taken 6 times.
✗ Branch 4 not taken.
✓ Branch 6 taken 6 times.
✗ Branch 7 not taken.
✓ Branch 8 taken 6 times.
✗ Branch 9 not taken.
|
6 |
wire.gid_index() < 0 || wire.active_mtu() < IBV_MTU_256 || |
| 142 |
2/4
✓ Branch 1 taken 6 times.
✗ Branch 2 not taken.
✓ Branch 3 taken 6 times.
✗ Branch 4 not taken.
|
6 |
wire.active_mtu() > IBV_MTU_4096 || |
| 143 |
3/4
✓ Branch 1 taken 6 times.
✗ Branch 2 not taken.
✓ Branch 3 taken 2 times.
✓ Branch 4 taken 4 times.
|
6 |
(wire.link_layer() != IBV_LINK_LAYER_INFINIBAND && |
| 144 |
2/4
✓ Branch 1 taken 2 times.
✗ Branch 2 not taken.
✗ Branch 3 not taken.
✓ Branch 4 taken 2 times.
|
2 |
wire.link_layer() != IBV_LINK_LAYER_ETHERNET) || |
| 145 |
2/4
✓ Branch 1 taken 4 times.
✗ Branch 2 not taken.
✓ Branch 3 taken 4 times.
✗ Branch 4 not taken.
|
4 |
(wire.fabric_mode() == |
| 146 |
|
4 |
static_cast<std::uint32_t>(recstore::RdmaFabricMode::kIb) && |
| 147 |
2/4
✓ Branch 1 taken 4 times.
✗ Branch 2 not taken.
✓ Branch 3 taken 4 times.
✗ Branch 4 not taken.
|
4 |
wire.link_layer() != IBV_LINK_LAYER_INFINIBAND) || |
| 148 |
2/4
✓ Branch 1 taken 4 times.
✗ Branch 2 not taken.
✗ Branch 3 not taken.
✓ Branch 4 taken 4 times.
|
4 |
(wire.fabric_mode() != |
| 149 |
|
✗ |
static_cast<std::uint32_t>(recstore::RdmaFabricMode::kIb) && |
| 150 |
3/8
✓ Branch 0 taken 12 times.
✗ Branch 1 not taken.
✗ Branch 3 not taken.
✗ Branch 4 not taken.
✗ Branch 5 not taken.
✗ Branch 6 not taken.
✓ Branch 7 taken 8 times.
✓ Branch 8 taken 4 times.
|
24 |
wire.link_layer() != IBV_LINK_LAYER_ETHERNET) || |
| 151 |
2/4
✓ Branch 1 taken 4 times.
✗ Branch 2 not taken.
✗ Branch 3 not taken.
✓ Branch 4 taken 4 times.
|
4 |
wire.fabric_mode() > |
| 152 |
|
|
static_cast<std::uint32_t>(recstore::RdmaFabricMode::kRocEv2)) { |
| 153 |
1/2
✓ Branch 2 taken 8 times.
✗ Branch 3 not taken.
|
8 |
throw std::runtime_error("invalid versioned RawVerbsNodeMeta payload"); |
| 154 |
|
|
} |
| 155 |
|
4 |
RawVerbsNodeMeta meta{}; |
| 156 |
1/2
✓ Branch 1 taken 4 times.
✗ Branch 2 not taken.
|
4 |
meta.protocol_version = wire.protocol_version(); |
| 157 |
2/4
✓ Branch 1 taken 4 times.
✗ Branch 2 not taken.
✓ Branch 4 taken 4 times.
✗ Branch 5 not taken.
|
4 |
meta.deployment_id = wire.deployment_id(); |
| 158 |
1/2
✓ Branch 1 taken 4 times.
✗ Branch 2 not taken.
|
4 |
meta.deployment_epoch = wire.deployment_epoch(); |
| 159 |
2/4
✓ Branch 1 taken 4 times.
✗ Branch 2 not taken.
✓ Branch 4 taken 4 times.
✗ Branch 5 not taken.
|
4 |
meta.configuration_digest = wire.configuration_digest(); |
| 160 |
2/4
✓ Branch 1 taken 4 times.
✗ Branch 2 not taken.
✓ Branch 4 taken 4 times.
✗ Branch 5 not taken.
|
4 |
meta.fabric_digest = wire.fabric_digest(); |
| 161 |
1/2
✓ Branch 1 taken 4 times.
✗ Branch 2 not taken.
|
4 |
meta.node_id = static_cast<std::uint16_t>(wire.node_id()); |
| 162 |
1/2
✓ Branch 1 taken 4 times.
✗ Branch 2 not taken.
|
4 |
meta.logical_id = wire.logical_id(); |
| 163 |
1/2
✓ Branch 1 taken 4 times.
✗ Branch 2 not taken.
|
4 |
meta.node_role = static_cast<std::uint8_t>(wire.node_role()); |
| 164 |
1/2
✓ Branch 1 taken 4 times.
✗ Branch 2 not taken.
|
4 |
meta.lid = static_cast<std::uint16_t>(wire.lid()); |
| 165 |
1/2
✓ Branch 1 taken 4 times.
✗ Branch 2 not taken.
|
4 |
meta.qpn = wire.qpn(); |
| 166 |
1/2
✓ Branch 1 taken 4 times.
✗ Branch 2 not taken.
|
4 |
meta.psn = wire.psn(); |
| 167 |
1/2
✓ Branch 1 taken 4 times.
✗ Branch 2 not taken.
|
4 |
meta.rkey = wire.rkey(); |
| 168 |
1/2
✓ Branch 1 taken 4 times.
✗ Branch 2 not taken.
|
4 |
meta.base_addr = wire.base_addr(); |
| 169 |
1/2
✓ Branch 1 taken 4 times.
✗ Branch 2 not taken.
|
4 |
std::memcpy(meta.gid, wire.gid().data(), sizeof(meta.gid)); |
| 170 |
1/2
✓ Branch 1 taken 4 times.
✗ Branch 2 not taken.
|
4 |
meta.port_num = static_cast<std::uint8_t>(wire.port_num()); |
| 171 |
1/2
✓ Branch 1 taken 4 times.
✗ Branch 2 not taken.
|
4 |
meta.link_layer = static_cast<std::uint8_t>(wire.link_layer()); |
| 172 |
1/2
✓ Branch 1 taken 4 times.
✗ Branch 2 not taken.
|
4 |
meta.gid_index = wire.gid_index(); |
| 173 |
1/2
✓ Branch 1 taken 4 times.
✗ Branch 2 not taken.
|
4 |
meta.active_mtu = static_cast<std::uint8_t>(wire.active_mtu()); |
| 174 |
1/2
✓ Branch 1 taken 4 times.
✗ Branch 2 not taken.
|
4 |
meta.fabric_mode = static_cast<std::uint8_t>(wire.fabric_mode()); |
| 175 |
1/2
✓ Branch 1 taken 4 times.
✗ Branch 2 not taken.
|
4 |
meta.lane_generation = wire.lane_generation(); |
| 176 |
|
8 |
return meta; |
| 177 |
|
12 |
} |
| 178 |
|
|
|
| 179 |
|
8 |
void ValidateRawVerbsPeerMeta(const RawVerbsConfig& local, |
| 180 |
|
|
int expected_node_id, |
| 181 |
|
|
const RawVerbsNodeMeta& remote) { |
| 182 |
|
8 |
const bool expected_server = expected_node_id < local.num_servers; |
| 183 |
|
8 |
const int expected_logical_id = |
| 184 |
1/2
✓ Branch 0 taken 8 times.
✗ Branch 1 not taken.
|
8 |
expected_server ? expected_node_id : expected_node_id - local.num_servers; |
| 185 |
1/2
✓ Branch 0 taken 8 times.
✗ Branch 1 not taken.
|
8 |
const auto expected_role = static_cast<std::uint8_t>( |
| 186 |
|
|
expected_server ? recstore::RdmaNodeRole::kServer |
| 187 |
|
|
: recstore::RdmaNodeRole::kClient); |
| 188 |
|
8 |
const auto expected_link_layer = static_cast<std::uint8_t>( |
| 189 |
1/2
✓ Branch 0 taken 8 times.
✗ Branch 1 not taken.
|
8 |
local.fabric_mode == recstore::RdmaFabricMode::kIb |
| 190 |
|
|
? IBV_LINK_LAYER_INFINIBAND |
| 191 |
|
|
: IBV_LINK_LAYER_ETHERNET); |
| 192 |
|
12 |
const auto incompatible = [expected_node_id](const char* field) { |
| 193 |
1/2
✓ Branch 1 taken 6 times.
✗ Branch 2 not taken.
|
6 |
throw std::runtime_error( |
| 194 |
3/6
✓ Branch 2 taken 6 times.
✗ Branch 3 not taken.
✓ Branch 5 taken 6 times.
✗ Branch 6 not taken.
✓ Branch 8 taken 6 times.
✗ Branch 9 not taken.
|
12 |
"incompatible RDMA peer metadata field=" + std::string(field) + |
| 195 |
2/4
✓ Branch 2 taken 6 times.
✗ Branch 3 not taken.
✓ Branch 5 taken 6 times.
✗ Branch 6 not taken.
|
24 |
" node_id=" + std::to_string(expected_node_id)); |
| 196 |
|
8 |
}; |
| 197 |
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 8 times.
|
8 |
if (remote.node_id != expected_node_id) |
| 198 |
0/2
✗ Branch 1 not taken.
✗ Branch 2 not taken.
|
✗ |
incompatible("node_id"); |
| 199 |
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 8 times.
|
8 |
if (remote.logical_id != expected_logical_id) |
| 200 |
0/2
✗ Branch 1 not taken.
✗ Branch 2 not taken.
|
✗ |
incompatible("logical_id"); |
| 201 |
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 8 times.
|
8 |
if (remote.node_role != expected_role) |
| 202 |
0/2
✗ Branch 1 not taken.
✗ Branch 2 not taken.
|
✗ |
incompatible("node_role"); |
| 203 |
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 8 times.
|
8 |
if (remote.protocol_version != local.protocol_version) |
| 204 |
0/2
✗ Branch 1 not taken.
✗ Branch 2 not taken.
|
✗ |
incompatible("protocol_version"); |
| 205 |
1/2
✗ Branch 1 not taken.
✓ Branch 2 taken 8 times.
|
8 |
if (remote.deployment_id != local.deployment_id) |
| 206 |
0/2
✗ Branch 1 not taken.
✗ Branch 2 not taken.
|
✗ |
incompatible("deployment_id"); |
| 207 |
2/2
✓ Branch 0 taken 2 times.
✓ Branch 1 taken 6 times.
|
8 |
if (remote.deployment_epoch != local.deployment_epoch) |
| 208 |
1/2
✗ Branch 1 not taken.
✓ Branch 2 taken 2 times.
|
2 |
incompatible("deployment_epoch"); |
| 209 |
1/2
✗ Branch 1 not taken.
✓ Branch 2 taken 6 times.
|
6 |
if (remote.configuration_digest != local.configuration_digest) |
| 210 |
0/2
✗ Branch 1 not taken.
✗ Branch 2 not taken.
|
✗ |
incompatible("configuration_digest"); |
| 211 |
1/2
✗ Branch 1 not taken.
✓ Branch 2 taken 6 times.
|
6 |
if (remote.fabric_digest != local.fabric_digest) |
| 212 |
0/2
✗ Branch 1 not taken.
✗ Branch 2 not taken.
|
✗ |
incompatible("fabric_digest"); |
| 213 |
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 6 times.
|
6 |
if (remote.link_layer != expected_link_layer) |
| 214 |
0/2
✗ Branch 1 not taken.
✗ Branch 2 not taken.
|
✗ |
incompatible("link_layer"); |
| 215 |
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 6 times.
|
6 |
if (remote.fabric_mode != static_cast<std::uint8_t>(local.fabric_mode)) |
| 216 |
0/2
✗ Branch 1 not taken.
✗ Branch 2 not taken.
|
✗ |
incompatible("fabric_mode"); |
| 217 |
2/2
✓ Branch 0 taken 2 times.
✓ Branch 1 taken 4 times.
|
6 |
if (remote.port_num == 0) |
| 218 |
1/2
✗ Branch 1 not taken.
✓ Branch 2 taken 2 times.
|
2 |
incompatible("port_num"); |
| 219 |
2/2
✓ Branch 0 taken 2 times.
✓ Branch 1 taken 2 times.
|
4 |
if (remote.gid_index < 0) |
| 220 |
1/2
✗ Branch 1 not taken.
✓ Branch 2 taken 2 times.
|
2 |
incompatible("gid_index"); |
| 221 |
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 2 times.
|
2 |
if (remote.active_mtu == 0) |
| 222 |
0/2
✗ Branch 1 not taken.
✗ Branch 2 not taken.
|
✗ |
incompatible("active_mtu"); |
| 223 |
|
2 |
} |
| 224 |
|
|
|
| 225 |
|
|
class RdmaControlPlaneService final : public RdmaControlPlane::Service { |
| 226 |
|
|
public: |
| 227 |
|
26 |
explicit RdmaControlPlaneService(RdmaControlPlaneServer* owner) |
| 228 |
|
26 |
: owner_(owner) {} |
| 229 |
|
|
|
| 230 |
|
90 |
grpc::Status ValidateDeploymentIdentity( |
| 231 |
|
|
const std::string& deployment_id, std::uint64_t deployment_epoch) const { |
| 232 |
3/6
✓ Branch 1 taken 90 times.
✗ Branch 2 not taken.
✗ Branch 3 not taken.
✓ Branch 4 taken 90 times.
✗ Branch 5 not taken.
✓ Branch 6 taken 90 times.
|
90 |
if (deployment_id.empty() || deployment_epoch == 0) { |
| 233 |
|
|
return grpc::Status(grpc::StatusCode::INVALID_ARGUMENT, |
| 234 |
0/4
✗ Branch 2 not taken.
✗ Branch 3 not taken.
✗ Branch 5 not taken.
✗ Branch 6 not taken.
|
✗ |
"control-plane deployment identity is required"); |
| 235 |
|
|
} |
| 236 |
4/4
✓ Branch 1 taken 80 times.
✓ Branch 2 taken 10 times.
✓ Branch 3 taken 12 times.
✓ Branch 4 taken 78 times.
|
170 |
if (deployment_id != owner_->endpoint_.deployment_id || |
| 237 |
2/2
✓ Branch 0 taken 2 times.
✓ Branch 1 taken 78 times.
|
80 |
deployment_epoch != owner_->endpoint_.deployment_epoch) { |
| 238 |
|
|
return grpc::Status(grpc::StatusCode::FAILED_PRECONDITION, |
| 239 |
2/4
✓ Branch 2 taken 12 times.
✗ Branch 3 not taken.
✓ Branch 5 taken 12 times.
✗ Branch 6 not taken.
|
12 |
"control-plane deployment identity mismatch"); |
| 240 |
|
|
} |
| 241 |
|
78 |
return grpc::Status::OK; |
| 242 |
|
|
} |
| 243 |
|
|
|
| 244 |
|
4 |
grpc::Status PublishMeta(grpc::ServerContext*, |
| 245 |
|
|
const PublishMetaRequest* request, |
| 246 |
|
|
PublishMetaResponse*) override { |
| 247 |
|
|
const grpc::Status identity = ValidateDeploymentIdentity( |
| 248 |
3/6
✓ Branch 1 taken 4 times.
✗ Branch 2 not taken.
✓ Branch 4 taken 4 times.
✗ Branch 5 not taken.
✓ Branch 7 taken 4 times.
✗ Branch 8 not taken.
|
4 |
request->deployment_id(), request->deployment_epoch()); |
| 249 |
2/2
✓ Branch 1 taken 2 times.
✓ Branch 2 taken 2 times.
|
4 |
if (!identity.ok()) { |
| 250 |
1/2
✓ Branch 1 taken 2 times.
✗ Branch 2 not taken.
|
2 |
return identity; |
| 251 |
|
|
} |
| 252 |
3/6
✓ Branch 1 taken 2 times.
✗ Branch 2 not taken.
✓ Branch 4 taken 2 times.
✗ Branch 5 not taken.
✓ Branch 6 taken 2 times.
✗ Branch 7 not taken.
|
4 |
if (request->publisher_node_id() < 0 || request->publisher_lane() < 0 || |
| 253 |
6/12
✓ Branch 0 taken 2 times.
✗ Branch 1 not taken.
✓ Branch 3 taken 2 times.
✗ Branch 4 not taken.
✓ Branch 5 taken 2 times.
✗ Branch 6 not taken.
✓ Branch 8 taken 2 times.
✗ Branch 9 not taken.
✗ Branch 10 not taken.
✓ Branch 11 taken 2 times.
✗ Branch 12 not taken.
✓ Branch 13 taken 2 times.
|
4 |
request->receiver_node_id() < 0 || request->receiver_lane() < 0) { |
| 254 |
|
|
return grpc::Status(grpc::StatusCode::INVALID_ARGUMENT, |
| 255 |
0/4
✗ Branch 2 not taken.
✗ Branch 3 not taken.
✗ Branch 5 not taken.
✗ Branch 6 not taken.
|
✗ |
"metadata node and lane ids must be non-negative"); |
| 256 |
|
|
} |
| 257 |
|
|
const RdmaControlPlaneServer::MetaKey key{ |
| 258 |
1/2
✓ Branch 1 taken 2 times.
✗ Branch 2 not taken.
|
2 |
request->deployment_id(), |
| 259 |
1/2
✓ Branch 1 taken 2 times.
✗ Branch 2 not taken.
|
2 |
request->deployment_epoch(), |
| 260 |
|
4 |
request->publisher_node_id(), |
| 261 |
|
4 |
request->publisher_lane(), |
| 262 |
|
4 |
request->receiver_node_id(), |
| 263 |
|
4 |
request->receiver_lane(), |
| 264 |
5/10
✓ Branch 1 taken 2 times.
✗ Branch 2 not taken.
✓ Branch 4 taken 2 times.
✗ Branch 5 not taken.
✓ Branch 7 taken 2 times.
✗ Branch 8 not taken.
✓ Branch 10 taken 2 times.
✗ Branch 11 not taken.
✓ Branch 13 taken 2 times.
✗ Branch 14 not taken.
|
4 |
}; |
| 265 |
|
2 |
RawVerbsNodeMeta meta; |
| 266 |
|
|
try { |
| 267 |
2/4
✓ Branch 1 taken 2 times.
✗ Branch 2 not taken.
✓ Branch 4 taken 2 times.
✗ Branch 5 not taken.
|
2 |
meta = DecodeMetaBytes(request->meta()); |
| 268 |
0/2
✗ Branch 0 not taken.
✗ Branch 1 not taken.
|
✗ |
} catch (const std::exception& error) { |
| 269 |
0/4
✗ Branch 3 not taken.
✗ Branch 4 not taken.
✗ Branch 6 not taken.
✗ Branch 7 not taken.
|
✗ |
return grpc::Status(grpc::StatusCode::INVALID_ARGUMENT, error.what()); |
| 270 |
|
✗ |
} |
| 271 |
1/2
✓ Branch 0 taken 2 times.
✗ Branch 1 not taken.
|
2 |
if (meta.protocol_version != owner_->endpoint_.protocol_version || |
| 272 |
1/2
✓ Branch 1 taken 2 times.
✗ Branch 2 not taken.
|
4 |
meta.configuration_digest != owner_->endpoint_.configuration_digest || |
| 273 |
1/2
✓ Branch 1 taken 2 times.
✗ Branch 2 not taken.
|
4 |
meta.fabric_digest != owner_->endpoint_.fabric_digest || |
| 274 |
1/2
✓ Branch 1 taken 2 times.
✗ Branch 2 not taken.
|
2 |
meta.deployment_id != request->deployment_id() || |
| 275 |
4/8
✓ Branch 0 taken 2 times.
✗ Branch 1 not taken.
✓ Branch 3 taken 2 times.
✗ Branch 4 not taken.
✓ Branch 5 taken 2 times.
✗ Branch 6 not taken.
✗ Branch 7 not taken.
✓ Branch 8 taken 2 times.
|
6 |
meta.deployment_epoch != request->deployment_epoch() || |
| 276 |
2/4
✓ Branch 1 taken 2 times.
✗ Branch 2 not taken.
✗ Branch 3 not taken.
✓ Branch 4 taken 2 times.
|
2 |
meta.node_id != request->publisher_node_id()) { |
| 277 |
|
|
return grpc::Status(grpc::StatusCode::INVALID_ARGUMENT, |
| 278 |
0/4
✗ Branch 2 not taken.
✗ Branch 3 not taken.
✗ Branch 5 not taken.
✗ Branch 6 not taken.
|
✗ |
"metadata deployment identity mismatch"); |
| 279 |
|
|
} |
| 280 |
|
|
{ |
| 281 |
1/2
✓ Branch 1 taken 2 times.
✗ Branch 2 not taken.
|
2 |
std::lock_guard<std::mutex> guard(owner_->mu_); |
| 282 |
2/4
✓ Branch 1 taken 2 times.
✗ Branch 2 not taken.
✓ Branch 4 taken 2 times.
✗ Branch 5 not taken.
|
2 |
owner_->metadata_[key] = meta; |
| 283 |
|
2 |
} |
| 284 |
|
2 |
owner_->cv_.notify_all(); |
| 285 |
1/2
✓ Branch 1 taken 2 times.
✗ Branch 2 not taken.
|
2 |
return grpc::Status::OK; |
| 286 |
|
4 |
} |
| 287 |
|
|
|
| 288 |
|
4 |
grpc::Status GetMeta(grpc::ServerContext*, |
| 289 |
|
|
const GetMetaRequest* request, |
| 290 |
|
|
GetMetaResponse* response) override { |
| 291 |
|
|
const grpc::Status identity = ValidateDeploymentIdentity( |
| 292 |
3/6
✓ Branch 1 taken 4 times.
✗ Branch 2 not taken.
✓ Branch 4 taken 4 times.
✗ Branch 5 not taken.
✓ Branch 7 taken 4 times.
✗ Branch 8 not taken.
|
4 |
request->deployment_id(), request->deployment_epoch()); |
| 293 |
2/2
✓ Branch 1 taken 2 times.
✓ Branch 2 taken 2 times.
|
4 |
if (!identity.ok()) { |
| 294 |
1/2
✓ Branch 1 taken 2 times.
✗ Branch 2 not taken.
|
2 |
return identity; |
| 295 |
|
|
} |
| 296 |
3/6
✓ Branch 1 taken 2 times.
✗ Branch 2 not taken.
✓ Branch 4 taken 2 times.
✗ Branch 5 not taken.
✓ Branch 6 taken 2 times.
✗ Branch 7 not taken.
|
4 |
if (request->publisher_node_id() < 0 || request->publisher_lane() < 0 || |
| 297 |
6/12
✓ Branch 0 taken 2 times.
✗ Branch 1 not taken.
✓ Branch 3 taken 2 times.
✗ Branch 4 not taken.
✓ Branch 5 taken 2 times.
✗ Branch 6 not taken.
✓ Branch 8 taken 2 times.
✗ Branch 9 not taken.
✗ Branch 10 not taken.
✓ Branch 11 taken 2 times.
✗ Branch 12 not taken.
✓ Branch 13 taken 2 times.
|
4 |
request->receiver_node_id() < 0 || request->receiver_lane() < 0) { |
| 298 |
|
|
return grpc::Status(grpc::StatusCode::INVALID_ARGUMENT, |
| 299 |
0/4
✗ Branch 2 not taken.
✗ Branch 3 not taken.
✗ Branch 5 not taken.
✗ Branch 6 not taken.
|
✗ |
"metadata node and lane ids must be non-negative"); |
| 300 |
|
|
} |
| 301 |
|
|
const RdmaControlPlaneServer::MetaKey key{ |
| 302 |
1/2
✓ Branch 1 taken 2 times.
✗ Branch 2 not taken.
|
2 |
request->deployment_id(), |
| 303 |
1/2
✓ Branch 1 taken 2 times.
✗ Branch 2 not taken.
|
2 |
request->deployment_epoch(), |
| 304 |
|
4 |
request->publisher_node_id(), |
| 305 |
|
4 |
request->publisher_lane(), |
| 306 |
|
4 |
request->receiver_node_id(), |
| 307 |
|
4 |
request->receiver_lane(), |
| 308 |
5/10
✓ Branch 1 taken 2 times.
✗ Branch 2 not taken.
✓ Branch 4 taken 2 times.
✗ Branch 5 not taken.
✓ Branch 7 taken 2 times.
✗ Branch 8 not taken.
✓ Branch 10 taken 2 times.
✗ Branch 11 not taken.
✓ Branch 13 taken 2 times.
✗ Branch 14 not taken.
|
4 |
}; |
| 309 |
|
|
const int timeout_ms = |
| 310 |
1/2
✓ Branch 1 taken 2 times.
✗ Branch 2 not taken.
|
2 |
request->timeout_ms() > 0 |
| 311 |
2/4
✓ Branch 0 taken 2 times.
✗ Branch 1 not taken.
✓ Branch 3 taken 2 times.
✗ Branch 4 not taken.
|
2 |
? request->timeout_ms() |
| 312 |
|
✗ |
: owner_->endpoint_.timeout_ms; |
| 313 |
1/2
✓ Branch 1 taken 2 times.
✗ Branch 2 not taken.
|
2 |
std::unique_lock<std::mutex> lock(owner_->mu_); |
| 314 |
|
|
const bool ready = |
| 315 |
1/2
✓ Branch 2 taken 2 times.
✗ Branch 3 not taken.
|
2 |
owner_->cv_.wait_for(lock, std::chrono::milliseconds(timeout_ms), [&] { |
| 316 |
3/4
✓ Branch 1 taken 4 times.
✗ Branch 2 not taken.
✓ Branch 3 taken 2 times.
✓ Branch 4 taken 2 times.
|
8 |
return owner_->stop_requested_.load(std::memory_order_relaxed) || |
| 317 |
1/2
✓ Branch 2 taken 4 times.
✗ Branch 3 not taken.
|
8 |
owner_->metadata_.find(key) != owner_->metadata_.end(); |
| 318 |
|
|
}); |
| 319 |
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 2 times.
|
2 |
if (!ready) { |
| 320 |
|
|
return MakeDeadlineExceeded( |
| 321 |
0/6
✗ Branch 1 not taken.
✗ Branch 2 not taken.
✗ Branch 4 not taken.
✗ Branch 5 not taken.
✗ Branch 7 not taken.
✗ Branch 8 not taken.
|
✗ |
"get_meta timeout key=" + std::to_string(key.publisher_node_id) + |
| 322 |
0/8
✗ Branch 1 not taken.
✗ Branch 2 not taken.
✗ Branch 4 not taken.
✗ Branch 5 not taken.
✗ Branch 7 not taken.
✗ Branch 8 not taken.
✗ Branch 10 not taken.
✗ Branch 11 not taken.
|
✗ |
":" + std::to_string(key.publisher_lane) + "->" + |
| 323 |
0/6
✗ Branch 1 not taken.
✗ Branch 2 not taken.
✗ Branch 4 not taken.
✗ Branch 5 not taken.
✗ Branch 7 not taken.
✗ Branch 8 not taken.
|
✗ |
std::to_string(key.receiver_node_id) + ":" + |
| 324 |
0/4
✗ Branch 1 not taken.
✗ Branch 2 not taken.
✗ Branch 4 not taken.
✗ Branch 5 not taken.
|
✗ |
std::to_string(key.receiver_lane)); |
| 325 |
|
|
} |
| 326 |
1/2
✗ Branch 1 not taken.
✓ Branch 2 taken 2 times.
|
2 |
if (owner_->stop_requested_.load(std::memory_order_relaxed)) { |
| 327 |
0/4
✗ Branch 2 not taken.
✗ Branch 3 not taken.
✗ Branch 5 not taken.
✗ Branch 6 not taken.
|
✗ |
return MakeUnavailable("control-plane stopping"); |
| 328 |
|
|
} |
| 329 |
2/4
✓ Branch 1 taken 2 times.
✗ Branch 2 not taken.
✓ Branch 4 taken 2 times.
✗ Branch 5 not taken.
|
4 |
response->set_meta(EncodeMetaBytes(owner_->metadata_.at(key))); |
| 330 |
1/2
✓ Branch 1 taken 2 times.
✗ Branch 2 not taken.
|
2 |
return grpc::Status::OK; |
| 331 |
|
4 |
} |
| 332 |
|
|
|
| 333 |
|
24 |
grpc::Status PublishServerReady(grpc::ServerContext*, |
| 334 |
|
|
const PublishServerReadyRequest* request, |
| 335 |
|
|
PublishServerReadyResponse*) override { |
| 336 |
|
|
const grpc::Status identity = ValidateDeploymentIdentity( |
| 337 |
3/6
✓ Branch 1 taken 24 times.
✗ Branch 2 not taken.
✓ Branch 4 taken 24 times.
✗ Branch 5 not taken.
✓ Branch 7 taken 24 times.
✗ Branch 8 not taken.
|
24 |
request->deployment_id(), request->deployment_epoch()); |
| 338 |
2/2
✓ Branch 1 taken 2 times.
✓ Branch 2 taken 22 times.
|
24 |
if (!identity.ok()) { |
| 339 |
1/2
✓ Branch 1 taken 2 times.
✗ Branch 2 not taken.
|
2 |
return identity; |
| 340 |
|
|
} |
| 341 |
2/4
✓ Branch 1 taken 22 times.
✗ Branch 2 not taken.
✓ Branch 3 taken 22 times.
✗ Branch 4 not taken.
|
44 |
if (request->protocol_version() != owner_->endpoint_.protocol_version || |
| 342 |
2/4
✓ Branch 1 taken 22 times.
✗ Branch 2 not taken.
✓ Branch 4 taken 22 times.
✗ Branch 5 not taken.
|
44 |
request->configuration_digest().empty() || |
| 343 |
2/4
✓ Branch 1 taken 22 times.
✗ Branch 2 not taken.
✓ Branch 4 taken 22 times.
✗ Branch 5 not taken.
|
44 |
request->fabric_digest().empty() || |
| 344 |
|
22 |
request->configuration_digest() != |
| 345 |
2/4
✓ Branch 1 taken 22 times.
✗ Branch 2 not taken.
✓ Branch 3 taken 22 times.
✗ Branch 4 not taken.
|
44 |
owner_->endpoint_.configuration_digest || |
| 346 |
1/2
✓ Branch 1 taken 22 times.
✗ Branch 2 not taken.
|
22 |
request->fabric_digest() != owner_->endpoint_.fabric_digest || |
| 347 |
4/8
✓ Branch 0 taken 22 times.
✗ Branch 1 not taken.
✓ Branch 3 taken 22 times.
✗ Branch 4 not taken.
✓ Branch 5 taken 22 times.
✗ Branch 6 not taken.
✗ Branch 7 not taken.
✓ Branch 8 taken 22 times.
|
66 |
request->server_id() < 0 || |
| 348 |
2/2
✓ Branch 0 taken 8 times.
✓ Branch 1 taken 14 times.
|
22 |
(owner_->endpoint_.num_servers > 0 && |
| 349 |
2/4
✓ Branch 1 taken 8 times.
✗ Branch 2 not taken.
✗ Branch 3 not taken.
✓ Branch 4 taken 8 times.
|
8 |
request->server_id() >= owner_->endpoint_.num_servers)) { |
| 350 |
|
|
return grpc::Status(grpc::StatusCode::INVALID_ARGUMENT, |
| 351 |
0/4
✗ Branch 2 not taken.
✗ Branch 3 not taken.
✗ Branch 5 not taken.
✗ Branch 6 not taken.
|
✗ |
"invalid versioned server READY record"); |
| 352 |
|
|
} |
| 353 |
|
|
{ |
| 354 |
1/2
✓ Branch 1 taken 22 times.
✗ Branch 2 not taken.
|
22 |
std::lock_guard<std::mutex> guard(owner_->mu_); |
| 355 |
|
44 |
auto& record = owner_->ready_servers_[{request->deployment_id(), |
| 356 |
1/2
✓ Branch 1 taken 22 times.
✗ Branch 2 not taken.
|
22 |
request->deployment_epoch(), |
| 357 |
4/8
✓ Branch 1 taken 22 times.
✗ Branch 2 not taken.
✓ Branch 4 taken 22 times.
✗ Branch 5 not taken.
✓ Branch 7 taken 22 times.
✗ Branch 8 not taken.
✓ Branch 10 taken 22 times.
✗ Branch 11 not taken.
|
44 |
request->server_id()}]; |
| 358 |
2/2
✓ Branch 0 taken 2 times.
✓ Branch 1 taken 20 times.
|
22 |
if (record.state == |
| 359 |
|
|
RdmaControlPlaneServer::ReadyRecord::State::kDraining) { |
| 360 |
|
|
return grpc::Status( |
| 361 |
|
|
grpc::StatusCode::FAILED_PRECONDITION, |
| 362 |
2/4
✓ Branch 2 taken 2 times.
✗ Branch 3 not taken.
✓ Branch 5 taken 2 times.
✗ Branch 6 not taken.
|
2 |
"draining server cannot become ready in same epoch"); |
| 363 |
|
|
} |
| 364 |
1/2
✓ Branch 1 taken 20 times.
✗ Branch 2 not taken.
|
20 |
record.protocol_version = request->protocol_version(); |
| 365 |
2/4
✓ Branch 1 taken 20 times.
✗ Branch 2 not taken.
✓ Branch 4 taken 20 times.
✗ Branch 5 not taken.
|
20 |
record.configuration_digest = request->configuration_digest(); |
| 366 |
2/4
✓ Branch 1 taken 20 times.
✗ Branch 2 not taken.
✓ Branch 4 taken 20 times.
✗ Branch 5 not taken.
|
20 |
record.fabric_digest = request->fabric_digest(); |
| 367 |
|
20 |
record.state = RdmaControlPlaneServer::ReadyRecord::State::kServing; |
| 368 |
|
20 |
record.last_seen = std::chrono::steady_clock::now(); |
| 369 |
2/2
✓ Branch 1 taken 20 times.
✓ Branch 2 taken 2 times.
|
22 |
} |
| 370 |
|
20 |
owner_->cv_.notify_all(); |
| 371 |
1/2
✓ Branch 1 taken 20 times.
✗ Branch 2 not taken.
|
20 |
return grpc::Status::OK; |
| 372 |
|
24 |
} |
| 373 |
|
|
|
| 374 |
|
4 |
grpc::Status PublishServerDraining( |
| 375 |
|
|
grpc::ServerContext*, |
| 376 |
|
|
const PublishServerDrainingRequest* request, |
| 377 |
|
|
PublishServerDrainingResponse*) override { |
| 378 |
|
|
const grpc::Status identity = ValidateDeploymentIdentity( |
| 379 |
3/6
✓ Branch 1 taken 4 times.
✗ Branch 2 not taken.
✓ Branch 4 taken 4 times.
✗ Branch 5 not taken.
✓ Branch 7 taken 4 times.
✗ Branch 8 not taken.
|
4 |
request->deployment_id(), request->deployment_epoch()); |
| 380 |
1/2
✗ Branch 1 not taken.
✓ Branch 2 taken 4 times.
|
4 |
if (!identity.ok()) { |
| 381 |
0/2
✗ Branch 1 not taken.
✗ Branch 2 not taken.
|
✗ |
return identity; |
| 382 |
|
|
} |
| 383 |
2/4
✓ Branch 1 taken 4 times.
✗ Branch 2 not taken.
✓ Branch 3 taken 4 times.
✗ Branch 4 not taken.
|
8 |
if (request->protocol_version() != owner_->endpoint_.protocol_version || |
| 384 |
|
4 |
request->configuration_digest() != |
| 385 |
2/4
✓ Branch 1 taken 4 times.
✗ Branch 2 not taken.
✓ Branch 3 taken 4 times.
✗ Branch 4 not taken.
|
8 |
owner_->endpoint_.configuration_digest || |
| 386 |
1/2
✓ Branch 1 taken 4 times.
✗ Branch 2 not taken.
|
4 |
request->fabric_digest() != owner_->endpoint_.fabric_digest || |
| 387 |
4/8
✓ Branch 0 taken 4 times.
✗ Branch 1 not taken.
✓ Branch 3 taken 4 times.
✗ Branch 4 not taken.
✓ Branch 5 taken 4 times.
✗ Branch 6 not taken.
✗ Branch 7 not taken.
✓ Branch 8 taken 4 times.
|
12 |
request->server_id() < 0 || |
| 388 |
1/2
✓ Branch 0 taken 4 times.
✗ Branch 1 not taken.
|
4 |
(owner_->endpoint_.num_servers > 0 && |
| 389 |
2/4
✓ Branch 1 taken 4 times.
✗ Branch 2 not taken.
✗ Branch 3 not taken.
✓ Branch 4 taken 4 times.
|
4 |
request->server_id() >= owner_->endpoint_.num_servers)) { |
| 390 |
|
|
return grpc::Status(grpc::StatusCode::INVALID_ARGUMENT, |
| 391 |
0/4
✗ Branch 2 not taken.
✗ Branch 3 not taken.
✗ Branch 5 not taken.
✗ Branch 6 not taken.
|
✗ |
"invalid versioned server DRAINING record"); |
| 392 |
|
|
} |
| 393 |
|
|
{ |
| 394 |
1/2
✓ Branch 1 taken 4 times.
✗ Branch 2 not taken.
|
4 |
std::lock_guard<std::mutex> guard(owner_->mu_); |
| 395 |
|
|
const RdmaControlPlaneServer::ReadyKey key{ |
| 396 |
1/2
✓ Branch 1 taken 4 times.
✗ Branch 2 not taken.
|
4 |
request->deployment_id(), |
| 397 |
1/2
✓ Branch 1 taken 4 times.
✗ Branch 2 not taken.
|
4 |
request->deployment_epoch(), |
| 398 |
2/4
✓ Branch 1 taken 4 times.
✗ Branch 2 not taken.
✓ Branch 4 taken 4 times.
✗ Branch 5 not taken.
|
8 |
request->server_id()}; |
| 399 |
1/2
✓ Branch 1 taken 4 times.
✗ Branch 2 not taken.
|
4 |
const auto it = owner_->ready_servers_.find(key); |
| 400 |
2/2
✓ Branch 2 taken 2 times.
✓ Branch 3 taken 2 times.
|
4 |
if (it == owner_->ready_servers_.end()) { |
| 401 |
|
|
return grpc::Status(grpc::StatusCode::FAILED_PRECONDITION, |
| 402 |
2/4
✓ Branch 2 taken 2 times.
✗ Branch 3 not taken.
✓ Branch 5 taken 2 times.
✗ Branch 6 not taken.
|
2 |
"server must be ready before draining"); |
| 403 |
|
|
} |
| 404 |
2/4
✓ Branch 2 taken 2 times.
✗ Branch 3 not taken.
✓ Branch 4 taken 2 times.
✗ Branch 5 not taken.
|
4 |
if (it->second.protocol_version != request->protocol_version() || |
| 405 |
4/8
✓ Branch 0 taken 2 times.
✗ Branch 1 not taken.
✓ Branch 3 taken 2 times.
✗ Branch 4 not taken.
✗ Branch 7 not taken.
✓ Branch 8 taken 2 times.
✗ Branch 9 not taken.
✓ Branch 10 taken 2 times.
|
4 |
it->second.configuration_digest != request->configuration_digest() || |
| 406 |
1/2
✓ Branch 1 taken 2 times.
✗ Branch 2 not taken.
|
2 |
it->second.fabric_digest != request->fabric_digest()) { |
| 407 |
|
|
return grpc::Status(grpc::StatusCode::FAILED_PRECONDITION, |
| 408 |
0/4
✗ Branch 2 not taken.
✗ Branch 3 not taken.
✗ Branch 5 not taken.
✗ Branch 6 not taken.
|
✗ |
"server DRAINING deployment contract mismatch"); |
| 409 |
|
|
} |
| 410 |
|
2 |
it->second.state = RdmaControlPlaneServer::ReadyRecord::State::kDraining; |
| 411 |
4/4
✓ Branch 1 taken 2 times.
✓ Branch 2 taken 2 times.
✓ Branch 4 taken 2 times.
✓ Branch 5 taken 2 times.
|
6 |
} |
| 412 |
|
2 |
owner_->cv_.notify_all(); |
| 413 |
1/2
✓ Branch 1 taken 2 times.
✗ Branch 2 not taken.
|
2 |
return grpc::Status::OK; |
| 414 |
|
4 |
} |
| 415 |
|
|
|
| 416 |
|
12 |
grpc::Status WaitServer(grpc::ServerContext*, |
| 417 |
|
|
const WaitServerRequest* request, |
| 418 |
|
|
WaitServerResponse*) override { |
| 419 |
|
|
const grpc::Status identity = ValidateDeploymentIdentity( |
| 420 |
3/6
✓ Branch 1 taken 12 times.
✗ Branch 2 not taken.
✓ Branch 4 taken 12 times.
✗ Branch 5 not taken.
✓ Branch 7 taken 12 times.
✗ Branch 8 not taken.
|
12 |
request->deployment_id(), request->deployment_epoch()); |
| 421 |
1/2
✗ Branch 1 not taken.
✓ Branch 2 taken 12 times.
|
12 |
if (!identity.ok()) { |
| 422 |
0/2
✗ Branch 1 not taken.
✗ Branch 2 not taken.
|
✗ |
return identity; |
| 423 |
|
|
} |
| 424 |
3/4
✓ Branch 1 taken 12 times.
✗ Branch 2 not taken.
✓ Branch 3 taken 10 times.
✓ Branch 4 taken 2 times.
|
24 |
if (request->protocol_version() != owner_->endpoint_.protocol_version || |
| 425 |
|
12 |
request->configuration_digest() != |
| 426 |
2/4
✓ Branch 1 taken 12 times.
✗ Branch 2 not taken.
✓ Branch 3 taken 10 times.
✗ Branch 4 not taken.
|
22 |
owner_->endpoint_.configuration_digest || |
| 427 |
1/2
✓ Branch 1 taken 10 times.
✗ Branch 2 not taken.
|
10 |
request->fabric_digest() != owner_->endpoint_.fabric_digest || |
| 428 |
5/8
✓ Branch 0 taken 12 times.
✗ Branch 1 not taken.
✓ Branch 3 taken 10 times.
✗ Branch 4 not taken.
✓ Branch 5 taken 10 times.
✗ Branch 6 not taken.
✓ Branch 7 taken 2 times.
✓ Branch 8 taken 10 times.
|
34 |
request->server_id() < 0 || |
| 429 |
2/2
✓ Branch 0 taken 4 times.
✓ Branch 1 taken 6 times.
|
10 |
(owner_->endpoint_.num_servers > 0 && |
| 430 |
2/4
✓ Branch 1 taken 4 times.
✗ Branch 2 not taken.
✗ Branch 3 not taken.
✓ Branch 4 taken 4 times.
|
4 |
request->server_id() >= owner_->endpoint_.num_servers)) { |
| 431 |
|
|
return grpc::Status(grpc::StatusCode::FAILED_PRECONDITION, |
| 432 |
2/4
✓ Branch 2 taken 2 times.
✗ Branch 3 not taken.
✓ Branch 5 taken 2 times.
✗ Branch 6 not taken.
|
2 |
"server READY deployment contract mismatch"); |
| 433 |
|
|
} |
| 434 |
|
|
const int timeout_ms = |
| 435 |
1/2
✓ Branch 1 taken 10 times.
✗ Branch 2 not taken.
|
10 |
request->timeout_ms() > 0 |
| 436 |
2/4
✓ Branch 0 taken 10 times.
✗ Branch 1 not taken.
✓ Branch 3 taken 10 times.
✗ Branch 4 not taken.
|
10 |
? request->timeout_ms() |
| 437 |
|
✗ |
: owner_->endpoint_.timeout_ms; |
| 438 |
|
|
const RdmaControlPlaneServer::ReadyKey key{ |
| 439 |
1/2
✓ Branch 1 taken 10 times.
✗ Branch 2 not taken.
|
10 |
request->deployment_id(), |
| 440 |
1/2
✓ Branch 1 taken 10 times.
✗ Branch 2 not taken.
|
10 |
request->deployment_epoch(), |
| 441 |
2/4
✓ Branch 1 taken 10 times.
✗ Branch 2 not taken.
✓ Branch 4 taken 10 times.
✗ Branch 5 not taken.
|
20 |
request->server_id()}; |
| 442 |
1/2
✓ Branch 1 taken 10 times.
✗ Branch 2 not taken.
|
10 |
std::unique_lock<std::mutex> lock(owner_->mu_); |
| 443 |
|
|
const bool ready = |
| 444 |
1/2
✓ Branch 2 taken 10 times.
✗ Branch 3 not taken.
|
10 |
owner_->cv_.wait_for(lock, std::chrono::milliseconds(timeout_ms), [&] { |
| 445 |
4/4
✓ Branch 1 taken 10 times.
✓ Branch 2 taken 2 times.
✓ Branch 3 taken 8 times.
✓ Branch 4 taken 2 times.
|
22 |
return owner_->stop_requested_.load(std::memory_order_relaxed) || |
| 446 |
1/2
✓ Branch 1 taken 10 times.
✗ Branch 2 not taken.
|
10 |
owner_->ready_servers_.find(key) != |
| 447 |
|
22 |
owner_->ready_servers_.end(); |
| 448 |
|
|
}); |
| 449 |
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 10 times.
|
10 |
if (!ready) { |
| 450 |
0/2
✗ Branch 1 not taken.
✗ Branch 2 not taken.
|
✗ |
return MakeDeadlineExceeded("wait_server timeout server_id=" + |
| 451 |
0/6
✗ Branch 1 not taken.
✗ Branch 2 not taken.
✗ Branch 4 not taken.
✗ Branch 5 not taken.
✗ Branch 7 not taken.
✗ Branch 8 not taken.
|
✗ |
std::to_string(request->server_id())); |
| 452 |
|
|
} |
| 453 |
2/2
✓ Branch 1 taken 2 times.
✓ Branch 2 taken 8 times.
|
10 |
if (owner_->stop_requested_.load(std::memory_order_relaxed)) { |
| 454 |
2/4
✓ Branch 2 taken 2 times.
✗ Branch 3 not taken.
✓ Branch 5 taken 2 times.
✗ Branch 6 not taken.
|
2 |
return MakeUnavailable("control-plane stopping"); |
| 455 |
|
|
} |
| 456 |
1/2
✓ Branch 1 taken 8 times.
✗ Branch 2 not taken.
|
8 |
const auto& record = owner_->ready_servers_.at(key); |
| 457 |
2/4
✓ Branch 1 taken 8 times.
✗ Branch 2 not taken.
✓ Branch 3 taken 8 times.
✗ Branch 4 not taken.
|
16 |
if (record.protocol_version != request->protocol_version() || |
| 458 |
4/8
✓ Branch 0 taken 8 times.
✗ Branch 1 not taken.
✓ Branch 3 taken 8 times.
✗ Branch 4 not taken.
✗ Branch 6 not taken.
✓ Branch 7 taken 8 times.
✗ Branch 8 not taken.
✓ Branch 9 taken 8 times.
|
16 |
record.configuration_digest != request->configuration_digest() || |
| 459 |
1/2
✓ Branch 1 taken 8 times.
✗ Branch 2 not taken.
|
8 |
record.fabric_digest != request->fabric_digest()) { |
| 460 |
|
|
return grpc::Status(grpc::StatusCode::FAILED_PRECONDITION, |
| 461 |
0/4
✗ Branch 2 not taken.
✗ Branch 3 not taken.
✗ Branch 5 not taken.
✗ Branch 6 not taken.
|
✗ |
"server READY deployment contract mismatch"); |
| 462 |
|
|
} |
| 463 |
2/2
✓ Branch 0 taken 2 times.
✓ Branch 1 taken 6 times.
|
8 |
if (record.state == RdmaControlPlaneServer::ReadyRecord::State::kDraining) { |
| 464 |
|
|
return MakeUnavailable( |
| 465 |
4/8
✓ Branch 1 taken 2 times.
✗ Branch 2 not taken.
✓ Branch 4 taken 2 times.
✗ Branch 5 not taken.
✓ Branch 7 taken 2 times.
✗ Branch 8 not taken.
✓ Branch 10 taken 2 times.
✗ Branch 11 not taken.
|
4 |
"server draining server_id=" + std::to_string(request->server_id())); |
| 466 |
|
|
} |
| 467 |
1/2
✓ Branch 1 taken 6 times.
✗ Branch 2 not taken.
|
6 |
return grpc::Status::OK; |
| 468 |
|
12 |
} |
| 469 |
|
|
|
| 470 |
|
8 |
grpc::Status WaitServerReady(grpc::ServerContext*, |
| 471 |
|
|
const WaitServerReadyRequest* request, |
| 472 |
|
|
WaitServerReadyResponse*) override { |
| 473 |
|
|
const grpc::Status identity = ValidateDeploymentIdentity( |
| 474 |
3/6
✓ Branch 1 taken 8 times.
✗ Branch 2 not taken.
✓ Branch 4 taken 8 times.
✗ Branch 5 not taken.
✓ Branch 7 taken 8 times.
✗ Branch 8 not taken.
|
8 |
request->deployment_id(), request->deployment_epoch()); |
| 475 |
1/2
✗ Branch 1 not taken.
✓ Branch 2 taken 8 times.
|
8 |
if (!identity.ok()) { |
| 476 |
0/2
✗ Branch 1 not taken.
✗ Branch 2 not taken.
|
✗ |
return identity; |
| 477 |
|
|
} |
| 478 |
3/4
✓ Branch 1 taken 8 times.
✗ Branch 2 not taken.
✓ Branch 3 taken 6 times.
✓ Branch 4 taken 2 times.
|
16 |
if (request->protocol_version() != owner_->endpoint_.protocol_version || |
| 479 |
|
8 |
request->configuration_digest() != |
| 480 |
2/4
✓ Branch 1 taken 8 times.
✗ Branch 2 not taken.
✓ Branch 3 taken 6 times.
✗ Branch 4 not taken.
|
14 |
owner_->endpoint_.configuration_digest || |
| 481 |
1/2
✓ Branch 1 taken 6 times.
✗ Branch 2 not taken.
|
6 |
request->fabric_digest() != owner_->endpoint_.fabric_digest || |
| 482 |
5/8
✓ Branch 0 taken 8 times.
✗ Branch 1 not taken.
✓ Branch 3 taken 6 times.
✗ Branch 4 not taken.
✓ Branch 5 taken 6 times.
✗ Branch 6 not taken.
✓ Branch 7 taken 2 times.
✓ Branch 8 taken 6 times.
|
22 |
request->num_servers() <= 0 || |
| 483 |
2/2
✓ Branch 0 taken 2 times.
✓ Branch 1 taken 4 times.
|
6 |
(owner_->endpoint_.num_servers > 0 && |
| 484 |
2/4
✓ Branch 1 taken 2 times.
✗ Branch 2 not taken.
✗ Branch 3 not taken.
✓ Branch 4 taken 2 times.
|
2 |
request->num_servers() > owner_->endpoint_.num_servers)) { |
| 485 |
|
|
return grpc::Status(grpc::StatusCode::FAILED_PRECONDITION, |
| 486 |
2/4
✓ Branch 2 taken 2 times.
✗ Branch 3 not taken.
✓ Branch 5 taken 2 times.
✗ Branch 6 not taken.
|
2 |
"server READY deployment contract mismatch"); |
| 487 |
|
|
} |
| 488 |
|
|
const int timeout_ms = |
| 489 |
1/2
✓ Branch 1 taken 6 times.
✗ Branch 2 not taken.
|
6 |
request->timeout_ms() > 0 |
| 490 |
2/4
✓ Branch 0 taken 6 times.
✗ Branch 1 not taken.
✓ Branch 3 taken 6 times.
✗ Branch 4 not taken.
|
6 |
? request->timeout_ms() |
| 491 |
|
✗ |
: owner_->endpoint_.timeout_ms; |
| 492 |
1/2
✓ Branch 1 taken 6 times.
✗ Branch 2 not taken.
|
6 |
std::unique_lock<std::mutex> lock(owner_->mu_); |
| 493 |
|
|
const bool ready = |
| 494 |
1/2
✓ Branch 2 taken 6 times.
✗ Branch 3 not taken.
|
6 |
owner_->cv_.wait_for(lock, std::chrono::milliseconds(timeout_ms), [&] { |
| 495 |
|
10 |
return owner_->stop_requested_.load(std::memory_order_relaxed) || |
| 496 |
|
10 |
owner_->ReadyCount( |
| 497 |
|
10 |
request->deployment_id(), |
| 498 |
|
|
request->deployment_epoch(), |
| 499 |
|
|
request->protocol_version(), |
| 500 |
|
|
request->configuration_digest(), |
| 501 |
2/2
✓ Branch 1 taken 8 times.
✓ Branch 2 taken 2 times.
|
10 |
request->fabric_digest()) >= request->num_servers() || |
| 502 |
1/2
✓ Branch 6 taken 8 times.
✗ Branch 7 not taken.
|
8 |
owner_->HasReadyContractMismatch( |
| 503 |
|
|
request->deployment_id(), |
| 504 |
|
|
request->deployment_epoch(), |
| 505 |
|
|
request->protocol_version(), |
| 506 |
|
|
request->configuration_digest(), |
| 507 |
1/2
✓ Branch 0 taken 10 times.
✗ Branch 1 not taken.
|
20 |
request->fabric_digest()) || |
| 508 |
2/2
✓ Branch 4 taken 2 times.
✓ Branch 5 taken 6 times.
|
8 |
owner_->HasDrainingServer(request->deployment_id(), |
| 509 |
|
|
request->deployment_epoch(), |
| 510 |
|
10 |
request->num_servers()); |
| 511 |
|
|
}); |
| 512 |
7/14
✓ Branch 1 taken 6 times.
✗ Branch 2 not taken.
✓ Branch 4 taken 6 times.
✗ Branch 5 not taken.
✓ Branch 7 taken 6 times.
✗ Branch 8 not taken.
✓ Branch 10 taken 6 times.
✗ Branch 11 not taken.
✓ Branch 13 taken 6 times.
✗ Branch 14 not taken.
✓ Branch 16 taken 6 times.
✗ Branch 17 not taken.
✗ Branch 18 not taken.
✓ Branch 19 taken 6 times.
|
6 |
if (owner_->HasReadyContractMismatch( |
| 513 |
|
|
request->deployment_id(), |
| 514 |
|
|
request->deployment_epoch(), |
| 515 |
|
|
request->protocol_version(), |
| 516 |
|
|
request->configuration_digest(), |
| 517 |
|
|
request->fabric_digest())) { |
| 518 |
|
|
return grpc::Status(grpc::StatusCode::FAILED_PRECONDITION, |
| 519 |
0/4
✗ Branch 2 not taken.
✗ Branch 3 not taken.
✗ Branch 5 not taken.
✗ Branch 6 not taken.
|
✗ |
"server READY deployment contract mismatch"); |
| 520 |
|
|
} |
| 521 |
6/10
✓ Branch 1 taken 6 times.
✗ Branch 2 not taken.
✓ Branch 4 taken 6 times.
✗ Branch 5 not taken.
✓ Branch 7 taken 6 times.
✗ Branch 8 not taken.
✓ Branch 10 taken 6 times.
✗ Branch 11 not taken.
✓ Branch 12 taken 2 times.
✓ Branch 13 taken 4 times.
|
6 |
if (owner_->HasDrainingServer(request->deployment_id(), |
| 522 |
|
|
request->deployment_epoch(), |
| 523 |
|
|
request->num_servers())) { |
| 524 |
2/4
✓ Branch 2 taken 2 times.
✗ Branch 3 not taken.
✓ Branch 5 taken 2 times.
✗ Branch 6 not taken.
|
2 |
return MakeUnavailable("one or more RDMA servers are draining"); |
| 525 |
|
|
} |
| 526 |
2/2
✓ Branch 0 taken 2 times.
✓ Branch 1 taken 2 times.
|
4 |
if (!ready) { |
| 527 |
|
|
return MakeDeadlineExceeded( |
| 528 |
1/2
✓ Branch 1 taken 2 times.
✗ Branch 2 not taken.
|
2 |
"wait_server_ready timeout ready=" + |
| 529 |
7/14
✓ Branch 1 taken 2 times.
✗ Branch 2 not taken.
✓ Branch 4 taken 2 times.
✗ Branch 5 not taken.
✓ Branch 7 taken 2 times.
✗ Branch 8 not taken.
✓ Branch 10 taken 2 times.
✗ Branch 11 not taken.
✓ Branch 13 taken 2 times.
✗ Branch 14 not taken.
✓ Branch 16 taken 2 times.
✗ Branch 17 not taken.
✓ Branch 19 taken 2 times.
✗ Branch 20 not taken.
|
4 |
std::to_string(owner_->ReadyCount( |
| 530 |
|
|
request->deployment_id(), |
| 531 |
|
|
request->deployment_epoch(), |
| 532 |
|
|
request->protocol_version(), |
| 533 |
|
|
request->configuration_digest(), |
| 534 |
1/2
✓ Branch 1 taken 2 times.
✗ Branch 2 not taken.
|
4 |
request->fabric_digest())) + |
| 535 |
4/8
✓ Branch 1 taken 2 times.
✗ Branch 2 not taken.
✓ Branch 4 taken 2 times.
✗ Branch 5 not taken.
✓ Branch 7 taken 2 times.
✗ Branch 8 not taken.
✓ Branch 10 taken 2 times.
✗ Branch 11 not taken.
|
8 |
"/" + std::to_string(request->num_servers())); |
| 536 |
|
|
} |
| 537 |
1/2
✗ Branch 1 not taken.
✓ Branch 2 taken 2 times.
|
2 |
if (owner_->stop_requested_.load(std::memory_order_relaxed)) { |
| 538 |
0/4
✗ Branch 2 not taken.
✗ Branch 3 not taken.
✗ Branch 5 not taken.
✗ Branch 6 not taken.
|
✗ |
return MakeUnavailable("control-plane stopping"); |
| 539 |
|
|
} |
| 540 |
1/2
✓ Branch 1 taken 2 times.
✗ Branch 2 not taken.
|
2 |
return grpc::Status::OK; |
| 541 |
|
8 |
} |
| 542 |
|
|
|
| 543 |
|
22 |
grpc::Status Probe(grpc::ServerContext*, |
| 544 |
|
|
const ProbeRequest* request, |
| 545 |
|
|
ProbeResponse*) override { |
| 546 |
|
|
const grpc::Status identity = ValidateDeploymentIdentity( |
| 547 |
3/6
✓ Branch 1 taken 22 times.
✗ Branch 2 not taken.
✓ Branch 4 taken 22 times.
✗ Branch 5 not taken.
✓ Branch 7 taken 22 times.
✗ Branch 8 not taken.
|
22 |
request->deployment_id(), request->deployment_epoch()); |
| 548 |
2/2
✓ Branch 1 taken 4 times.
✓ Branch 2 taken 18 times.
|
22 |
if (!identity.ok()) { |
| 549 |
1/2
✓ Branch 1 taken 4 times.
✗ Branch 2 not taken.
|
4 |
return identity; |
| 550 |
|
|
} |
| 551 |
2/4
✓ Branch 1 taken 18 times.
✗ Branch 2 not taken.
✗ Branch 3 not taken.
✓ Branch 4 taken 18 times.
|
18 |
if (request->server_id() < 0) { |
| 552 |
|
|
return grpc::Status( |
| 553 |
0/4
✗ Branch 2 not taken.
✗ Branch 3 not taken.
✗ Branch 5 not taken.
✗ Branch 6 not taken.
|
✗ |
grpc::StatusCode::INVALID_ARGUMENT, "server id must be non-negative"); |
| 554 |
|
|
} |
| 555 |
1/2
✗ Branch 1 not taken.
✓ Branch 2 taken 18 times.
|
18 |
if (owner_->stop_requested_.load(std::memory_order_relaxed)) { |
| 556 |
0/4
✗ Branch 2 not taken.
✗ Branch 3 not taken.
✗ Branch 5 not taken.
✗ Branch 6 not taken.
|
✗ |
return MakeUnavailable("control-plane stopping"); |
| 557 |
|
|
} |
| 558 |
|
|
// Only a shard that already published READY has a liveness record; a |
| 559 |
|
|
// heartbeat never creates one, and READY remains the only way to serve. |
| 560 |
2/4
✓ Branch 1 taken 18 times.
✗ Branch 2 not taken.
✓ Branch 4 taken 18 times.
✗ Branch 5 not taken.
|
18 |
owner_->RecordHeartbeat(request->server_id()); |
| 561 |
1/2
✓ Branch 1 taken 18 times.
✗ Branch 2 not taken.
|
18 |
return grpc::Status::OK; |
| 562 |
|
22 |
} |
| 563 |
|
|
|
| 564 |
|
12 |
grpc::Status GetServerHealth(grpc::ServerContext*, |
| 565 |
|
|
const GetServerHealthRequest* request, |
| 566 |
|
|
GetServerHealthResponse* response) override { |
| 567 |
|
|
const grpc::Status identity = ValidateDeploymentIdentity( |
| 568 |
3/6
✓ Branch 1 taken 12 times.
✗ Branch 2 not taken.
✓ Branch 4 taken 12 times.
✗ Branch 5 not taken.
✓ Branch 7 taken 12 times.
✗ Branch 8 not taken.
|
12 |
request->deployment_id(), request->deployment_epoch()); |
| 569 |
2/2
✓ Branch 1 taken 2 times.
✓ Branch 2 taken 10 times.
|
12 |
if (!identity.ok()) { |
| 570 |
1/2
✓ Branch 1 taken 2 times.
✗ Branch 2 not taken.
|
2 |
return identity; |
| 571 |
|
|
} |
| 572 |
2/4
✓ Branch 1 taken 10 times.
✗ Branch 2 not taken.
✗ Branch 3 not taken.
✓ Branch 4 taken 10 times.
|
10 |
if (request->server_id() < 0) { |
| 573 |
|
|
return grpc::Status( |
| 574 |
0/4
✗ Branch 2 not taken.
✗ Branch 3 not taken.
✗ Branch 5 not taken.
✗ Branch 6 not taken.
|
✗ |
grpc::StatusCode::INVALID_ARGUMENT, "server id must be non-negative"); |
| 575 |
|
|
} |
| 576 |
1/2
✗ Branch 1 not taken.
✓ Branch 2 taken 10 times.
|
10 |
if (owner_->stop_requested_.load(std::memory_order_relaxed)) { |
| 577 |
0/4
✗ Branch 2 not taken.
✗ Branch 3 not taken.
✗ Branch 5 not taken.
✗ Branch 6 not taken.
|
✗ |
return MakeUnavailable("control-plane stopping"); |
| 578 |
|
|
} |
| 579 |
|
|
const RdmaServerHealthSnapshot snapshot = |
| 580 |
2/4
✓ Branch 1 taken 10 times.
✗ Branch 2 not taken.
✓ Branch 4 taken 10 times.
✗ Branch 5 not taken.
|
10 |
owner_->ServerHealthSnapshot(request->server_id()); |
| 581 |
1/2
✓ Branch 1 taken 10 times.
✗ Branch 2 not taken.
|
10 |
response->set_known(snapshot.known); |
| 582 |
1/2
✓ Branch 1 taken 10 times.
✗ Branch 2 not taken.
|
10 |
response->set_state( |
| 583 |
2/2
✓ Branch 0 taken 8 times.
✓ Branch 1 taken 2 times.
|
10 |
snapshot.known ? static_cast<std::int32_t>(snapshot.health) : -1); |
| 584 |
1/2
✓ Branch 1 taken 10 times.
✗ Branch 2 not taken.
|
10 |
response->set_last_seen_age_ms( |
| 585 |
2/2
✓ Branch 0 taken 8 times.
✓ Branch 1 taken 2 times.
|
10 |
snapshot.known ? snapshot.last_seen_age_ms : -1); |
| 586 |
1/2
✓ Branch 1 taken 10 times.
✗ Branch 2 not taken.
|
10 |
return grpc::Status::OK; |
| 587 |
|
12 |
} |
| 588 |
|
|
|
| 589 |
|
|
private: |
| 590 |
|
|
RdmaControlPlaneServer* owner_; |
| 591 |
|
|
}; |
| 592 |
|
|
|
| 593 |
|
34 |
RdmaControlPlaneClient::RdmaControlPlaneClient( |
| 594 |
|
34 |
RdmaControlPlaneEndpoint endpoint) |
| 595 |
|
34 |
: endpoint_(std::move(endpoint)), |
| 596 |
1/2
✓ Branch 1 taken 34 times.
✗ Branch 2 not taken.
|
34 |
channels_(std::make_unique<RdmaControlPlaneChannelCache>(endpoint_)) {} |
| 597 |
|
|
|
| 598 |
|
|
// Out of line because the cache type is incomplete in the header. |
| 599 |
|
34 |
RdmaControlPlaneClient::~RdmaControlPlaneClient() = default; |
| 600 |
|
|
|
| 601 |
|
4 |
void RdmaControlPlaneClient::PublishMeta( |
| 602 |
|
|
int publisher_node_id, |
| 603 |
|
|
int publisher_lane, |
| 604 |
|
|
int receiver_node_id, |
| 605 |
|
|
int receiver_lane, |
| 606 |
|
|
const RawVerbsNodeMeta& meta) const { |
| 607 |
|
4 |
RdmaControlPlane::Stub* stub = channels_->stub.get(); |
| 608 |
|
|
|
| 609 |
1/2
✓ Branch 1 taken 4 times.
✗ Branch 2 not taken.
|
4 |
PublishMetaRequest request; |
| 610 |
1/2
✓ Branch 1 taken 4 times.
✗ Branch 2 not taken.
|
4 |
request.set_publisher_node_id(publisher_node_id); |
| 611 |
1/2
✓ Branch 1 taken 4 times.
✗ Branch 2 not taken.
|
4 |
request.set_publisher_lane(publisher_lane); |
| 612 |
1/2
✓ Branch 1 taken 4 times.
✗ Branch 2 not taken.
|
4 |
request.set_receiver_node_id(receiver_node_id); |
| 613 |
1/2
✓ Branch 1 taken 4 times.
✗ Branch 2 not taken.
|
4 |
request.set_receiver_lane(receiver_lane); |
| 614 |
1/2
✓ Branch 1 taken 4 times.
✗ Branch 2 not taken.
|
8 |
request.set_meta(EncodeMetaBytes(meta)); |
| 615 |
1/2
✓ Branch 1 taken 4 times.
✗ Branch 2 not taken.
|
4 |
request.set_deployment_id(endpoint_.deployment_id); |
| 616 |
1/2
✓ Branch 1 taken 4 times.
✗ Branch 2 not taken.
|
4 |
request.set_deployment_epoch(endpoint_.deployment_epoch); |
| 617 |
|
|
|
| 618 |
1/2
✓ Branch 1 taken 4 times.
✗ Branch 2 not taken.
|
4 |
PublishMetaResponse response; |
| 619 |
1/2
✓ Branch 1 taken 4 times.
✗ Branch 2 not taken.
|
4 |
grpc::ClientContext context; |
| 620 |
2/4
✓ Branch 1 taken 4 times.
✗ Branch 2 not taken.
✓ Branch 4 taken 4 times.
✗ Branch 5 not taken.
|
4 |
context.set_deadline(DeadlineFromNow(endpoint_.timeout_ms)); |
| 621 |
4/6
✓ Branch 2 taken 4 times.
✗ Branch 3 not taken.
✓ Branch 5 taken 4 times.
✗ Branch 6 not taken.
✓ Branch 8 taken 2 times.
✓ Branch 9 taken 2 times.
|
10 |
ThrowIfNotOk(stub->PublishMeta(&context, request, &response), "publish_meta"); |
| 622 |
|
8 |
} |
| 623 |
|
|
|
| 624 |
|
4 |
RawVerbsNodeMeta RdmaControlPlaneClient::GetMeta( |
| 625 |
|
|
int publisher_node_id, |
| 626 |
|
|
int publisher_lane, |
| 627 |
|
|
int receiver_node_id, |
| 628 |
|
|
int receiver_lane, |
| 629 |
|
|
int timeout_ms) const { |
| 630 |
|
4 |
const int effective_timeout_ms = |
| 631 |
1/2
✓ Branch 0 taken 4 times.
✗ Branch 1 not taken.
|
4 |
timeout_ms > 0 ? timeout_ms : endpoint_.timeout_ms; |
| 632 |
|
4 |
RdmaControlPlane::Stub* stub = channels_->stub.get(); |
| 633 |
|
|
|
| 634 |
1/2
✓ Branch 1 taken 4 times.
✗ Branch 2 not taken.
|
4 |
GetMetaRequest request; |
| 635 |
1/2
✓ Branch 1 taken 4 times.
✗ Branch 2 not taken.
|
4 |
request.set_publisher_node_id(publisher_node_id); |
| 636 |
1/2
✓ Branch 1 taken 4 times.
✗ Branch 2 not taken.
|
4 |
request.set_publisher_lane(publisher_lane); |
| 637 |
1/2
✓ Branch 1 taken 4 times.
✗ Branch 2 not taken.
|
4 |
request.set_receiver_node_id(receiver_node_id); |
| 638 |
1/2
✓ Branch 1 taken 4 times.
✗ Branch 2 not taken.
|
4 |
request.set_receiver_lane(receiver_lane); |
| 639 |
1/2
✓ Branch 1 taken 4 times.
✗ Branch 2 not taken.
|
4 |
request.set_timeout_ms(effective_timeout_ms); |
| 640 |
1/2
✓ Branch 1 taken 4 times.
✗ Branch 2 not taken.
|
4 |
request.set_deployment_id(endpoint_.deployment_id); |
| 641 |
1/2
✓ Branch 1 taken 4 times.
✗ Branch 2 not taken.
|
4 |
request.set_deployment_epoch(endpoint_.deployment_epoch); |
| 642 |
|
|
|
| 643 |
1/2
✓ Branch 1 taken 4 times.
✗ Branch 2 not taken.
|
4 |
GetMetaResponse response; |
| 644 |
1/2
✓ Branch 1 taken 4 times.
✗ Branch 2 not taken.
|
4 |
grpc::ClientContext context; |
| 645 |
2/4
✓ Branch 1 taken 4 times.
✗ Branch 2 not taken.
✓ Branch 4 taken 4 times.
✗ Branch 5 not taken.
|
4 |
context.set_deadline(DeadlineFromNow(effective_timeout_ms)); |
| 646 |
4/6
✓ Branch 2 taken 4 times.
✗ Branch 3 not taken.
✓ Branch 5 taken 4 times.
✗ Branch 6 not taken.
✓ Branch 8 taken 2 times.
✓ Branch 9 taken 2 times.
|
10 |
ThrowIfNotOk(stub->GetMeta(&context, request, &response), "get_meta"); |
| 647 |
2/4
✓ Branch 1 taken 2 times.
✗ Branch 2 not taken.
✓ Branch 4 taken 2 times.
✗ Branch 5 not taken.
|
4 |
return DecodeMetaBytes(response.meta()); |
| 648 |
|
8 |
} |
| 649 |
|
|
|
| 650 |
|
24 |
void RdmaControlPlaneClient::PublishServerReady(int server_id) const { |
| 651 |
|
24 |
RdmaControlPlane::Stub* stub = channels_->stub.get(); |
| 652 |
|
|
|
| 653 |
1/2
✓ Branch 1 taken 24 times.
✗ Branch 2 not taken.
|
24 |
PublishServerReadyRequest request; |
| 654 |
1/2
✓ Branch 1 taken 24 times.
✗ Branch 2 not taken.
|
24 |
request.set_server_id(server_id); |
| 655 |
1/2
✓ Branch 1 taken 24 times.
✗ Branch 2 not taken.
|
24 |
request.set_protocol_version(endpoint_.protocol_version); |
| 656 |
1/2
✓ Branch 1 taken 24 times.
✗ Branch 2 not taken.
|
24 |
request.set_deployment_id(endpoint_.deployment_id); |
| 657 |
1/2
✓ Branch 1 taken 24 times.
✗ Branch 2 not taken.
|
24 |
request.set_deployment_epoch(endpoint_.deployment_epoch); |
| 658 |
1/2
✓ Branch 1 taken 24 times.
✗ Branch 2 not taken.
|
24 |
request.set_configuration_digest(endpoint_.configuration_digest); |
| 659 |
1/2
✓ Branch 1 taken 24 times.
✗ Branch 2 not taken.
|
24 |
request.set_fabric_digest(endpoint_.fabric_digest); |
| 660 |
|
|
|
| 661 |
1/2
✓ Branch 1 taken 24 times.
✗ Branch 2 not taken.
|
24 |
PublishServerReadyResponse response; |
| 662 |
1/2
✓ Branch 1 taken 24 times.
✗ Branch 2 not taken.
|
24 |
grpc::ClientContext context; |
| 663 |
2/4
✓ Branch 1 taken 24 times.
✗ Branch 2 not taken.
✓ Branch 4 taken 24 times.
✗ Branch 5 not taken.
|
24 |
context.set_deadline(DeadlineFromNow(endpoint_.timeout_ms)); |
| 664 |
4/6
✓ Branch 2 taken 24 times.
✗ Branch 3 not taken.
✓ Branch 5 taken 24 times.
✗ Branch 6 not taken.
✓ Branch 8 taken 20 times.
✓ Branch 9 taken 4 times.
|
36 |
ThrowIfNotOk(stub->PublishServerReady(&context, request, &response), |
| 665 |
|
|
"server_ready"); |
| 666 |
|
32 |
} |
| 667 |
|
|
|
| 668 |
|
4 |
void RdmaControlPlaneClient::PublishServerDraining(int server_id) const { |
| 669 |
|
4 |
RdmaControlPlane::Stub* stub = channels_->stub.get(); |
| 670 |
|
|
|
| 671 |
1/2
✓ Branch 1 taken 4 times.
✗ Branch 2 not taken.
|
4 |
PublishServerDrainingRequest request; |
| 672 |
1/2
✓ Branch 1 taken 4 times.
✗ Branch 2 not taken.
|
4 |
request.set_server_id(server_id); |
| 673 |
1/2
✓ Branch 1 taken 4 times.
✗ Branch 2 not taken.
|
4 |
request.set_protocol_version(endpoint_.protocol_version); |
| 674 |
1/2
✓ Branch 1 taken 4 times.
✗ Branch 2 not taken.
|
4 |
request.set_deployment_id(endpoint_.deployment_id); |
| 675 |
1/2
✓ Branch 1 taken 4 times.
✗ Branch 2 not taken.
|
4 |
request.set_deployment_epoch(endpoint_.deployment_epoch); |
| 676 |
1/2
✓ Branch 1 taken 4 times.
✗ Branch 2 not taken.
|
4 |
request.set_configuration_digest(endpoint_.configuration_digest); |
| 677 |
1/2
✓ Branch 1 taken 4 times.
✗ Branch 2 not taken.
|
4 |
request.set_fabric_digest(endpoint_.fabric_digest); |
| 678 |
|
|
|
| 679 |
1/2
✓ Branch 1 taken 4 times.
✗ Branch 2 not taken.
|
4 |
PublishServerDrainingResponse response; |
| 680 |
1/2
✓ Branch 1 taken 4 times.
✗ Branch 2 not taken.
|
4 |
grpc::ClientContext context; |
| 681 |
2/4
✓ Branch 1 taken 4 times.
✗ Branch 2 not taken.
✓ Branch 4 taken 4 times.
✗ Branch 5 not taken.
|
4 |
context.set_deadline(DeadlineFromNow(endpoint_.timeout_ms)); |
| 682 |
4/6
✓ Branch 2 taken 4 times.
✗ Branch 3 not taken.
✓ Branch 5 taken 4 times.
✗ Branch 6 not taken.
✓ Branch 8 taken 2 times.
✓ Branch 9 taken 2 times.
|
10 |
ThrowIfNotOk(stub->PublishServerDraining(&context, request, &response), |
| 683 |
|
|
"server_draining"); |
| 684 |
|
8 |
} |
| 685 |
|
|
|
| 686 |
|
22 |
void RdmaControlPlaneClient::SendHeartbeat(int server_id) const { |
| 687 |
|
22 |
RdmaControlPlane::Stub* stub = channels_->stub.get(); |
| 688 |
|
|
|
| 689 |
1/2
✓ Branch 1 taken 22 times.
✗ Branch 2 not taken.
|
22 |
ProbeRequest request; |
| 690 |
1/2
✓ Branch 1 taken 22 times.
✗ Branch 2 not taken.
|
22 |
request.set_server_id(server_id); |
| 691 |
1/2
✓ Branch 1 taken 22 times.
✗ Branch 2 not taken.
|
22 |
request.set_deployment_id(endpoint_.deployment_id); |
| 692 |
1/2
✓ Branch 1 taken 22 times.
✗ Branch 2 not taken.
|
22 |
request.set_deployment_epoch(endpoint_.deployment_epoch); |
| 693 |
|
|
|
| 694 |
1/2
✓ Branch 1 taken 22 times.
✗ Branch 2 not taken.
|
22 |
ProbeResponse response; |
| 695 |
1/2
✓ Branch 1 taken 22 times.
✗ Branch 2 not taken.
|
22 |
grpc::ClientContext context; |
| 696 |
2/4
✓ Branch 1 taken 22 times.
✗ Branch 2 not taken.
✓ Branch 4 taken 22 times.
✗ Branch 5 not taken.
|
22 |
context.set_deadline(DeadlineFromNow(endpoint_.timeout_ms)); |
| 697 |
4/6
✓ Branch 2 taken 22 times.
✗ Branch 3 not taken.
✓ Branch 5 taken 22 times.
✗ Branch 6 not taken.
✓ Branch 8 taken 18 times.
✓ Branch 9 taken 4 times.
|
34 |
ThrowIfNotOk(stub->Probe(&context, request, &response), "probe"); |
| 698 |
|
30 |
} |
| 699 |
|
|
|
| 700 |
|
|
RdmaServerHealthSnapshot |
| 701 |
|
14 |
RdmaControlPlaneClient::GetServerHealth(int server_id, int timeout_ms) const { |
| 702 |
|
14 |
const int effective_timeout_ms = |
| 703 |
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 14 times.
|
14 |
timeout_ms > 0 ? timeout_ms : endpoint_.timeout_ms; |
| 704 |
|
14 |
RdmaServerHealthSnapshot snapshot; |
| 705 |
|
|
|
| 706 |
|
14 |
RdmaControlPlane::Stub* stub = channels_->stub.get(); |
| 707 |
|
|
|
| 708 |
1/2
✓ Branch 1 taken 14 times.
✗ Branch 2 not taken.
|
14 |
GetServerHealthRequest request; |
| 709 |
1/2
✓ Branch 1 taken 14 times.
✗ Branch 2 not taken.
|
14 |
request.set_server_id(server_id); |
| 710 |
1/2
✓ Branch 1 taken 14 times.
✗ Branch 2 not taken.
|
14 |
request.set_deployment_id(endpoint_.deployment_id); |
| 711 |
1/2
✓ Branch 1 taken 14 times.
✗ Branch 2 not taken.
|
14 |
request.set_deployment_epoch(endpoint_.deployment_epoch); |
| 712 |
|
|
|
| 713 |
1/2
✓ Branch 1 taken 14 times.
✗ Branch 2 not taken.
|
14 |
GetServerHealthResponse response; |
| 714 |
1/2
✓ Branch 1 taken 14 times.
✗ Branch 2 not taken.
|
14 |
grpc::ClientContext context; |
| 715 |
2/4
✓ Branch 1 taken 14 times.
✗ Branch 2 not taken.
✓ Branch 4 taken 14 times.
✗ Branch 5 not taken.
|
14 |
context.set_deadline(DeadlineFromNow(effective_timeout_ms)); |
| 716 |
|
|
const grpc::Status status = |
| 717 |
1/2
✓ Branch 1 taken 14 times.
✗ Branch 2 not taken.
|
14 |
stub->GetServerHealth(&context, request, &response); |
| 718 |
2/2
✓ Branch 1 taken 4 times.
✓ Branch 2 taken 10 times.
|
14 |
if (!status.ok()) { |
| 719 |
3/12
✓ Branch 0 taken 4 times.
✗ Branch 1 not taken.
✓ Branch 3 taken 4 times.
✗ Branch 4 not taken.
✗ Branch 5 not taken.
✓ Branch 6 taken 4 times.
✗ Branch 8 not taken.
✗ Branch 9 not taken.
✗ Branch 11 not taken.
✗ Branch 12 not taken.
✗ Branch 15 not taken.
✗ Branch 16 not taken.
|
4 |
VLOG(1) << "component=rdma_control_plane event=get_server_health_failed" |
| 720 |
0/4
✗ Branch 1 not taken.
✗ Branch 2 not taken.
✗ Branch 4 not taken.
✗ Branch 5 not taken.
|
✗ |
<< " server_id=" << server_id |
| 721 |
2/14
✗ Branch 1 not taken.
✗ Branch 2 not taken.
✗ Branch 4 not taken.
✗ Branch 5 not taken.
✗ Branch 7 not taken.
✗ Branch 8 not taken.
✗ Branch 10 not taken.
✗ Branch 11 not taken.
✗ Branch 13 not taken.
✓ Branch 14 taken 4 times.
✗ Branch 16 not taken.
✓ Branch 17 taken 4 times.
✗ Branch 18 not taken.
✗ Branch 19 not taken.
|
4 |
<< " status=" << GrpcStatusText(status); |
| 722 |
|
4 |
return snapshot; |
| 723 |
|
|
} |
| 724 |
3/4
✓ Branch 1 taken 10 times.
✗ Branch 2 not taken.
✓ Branch 3 taken 2 times.
✓ Branch 4 taken 8 times.
|
10 |
if (!response.known()) { |
| 725 |
|
2 |
return snapshot; |
| 726 |
|
|
} |
| 727 |
1/2
✓ Branch 1 taken 8 times.
✗ Branch 2 not taken.
|
8 |
const auto state = response.state(); |
| 728 |
2/4
✓ Branch 0 taken 8 times.
✗ Branch 1 not taken.
✗ Branch 2 not taken.
✓ Branch 3 taken 8 times.
|
8 |
if (state < static_cast<std::int32_t>(RdmaServerHealth::kServing) || |
| 729 |
|
|
state > static_cast<std::int32_t>(RdmaServerHealth::kDead)) { |
| 730 |
|
✗ |
return snapshot; |
| 731 |
|
|
} |
| 732 |
|
8 |
snapshot.known = true; |
| 733 |
|
8 |
snapshot.health = static_cast<RdmaServerHealth>(state); |
| 734 |
1/2
✓ Branch 1 taken 8 times.
✗ Branch 2 not taken.
|
8 |
snapshot.last_seen_age_ms = response.last_seen_age_ms(); |
| 735 |
|
8 |
return snapshot; |
| 736 |
|
14 |
} |
| 737 |
|
|
|
| 738 |
|
12 |
void RdmaControlPlaneClient::WaitServer(int server_id, int timeout_ms) const { |
| 739 |
|
12 |
const int effective_timeout_ms = |
| 740 |
1/2
✓ Branch 0 taken 12 times.
✗ Branch 1 not taken.
|
12 |
timeout_ms > 0 ? timeout_ms : endpoint_.timeout_ms; |
| 741 |
|
12 |
RdmaControlPlane::Stub* stub = channels_->stub.get(); |
| 742 |
|
|
|
| 743 |
1/2
✓ Branch 1 taken 12 times.
✗ Branch 2 not taken.
|
12 |
WaitServerRequest request; |
| 744 |
1/2
✓ Branch 1 taken 12 times.
✗ Branch 2 not taken.
|
12 |
request.set_server_id(server_id); |
| 745 |
1/2
✓ Branch 1 taken 12 times.
✗ Branch 2 not taken.
|
12 |
request.set_timeout_ms(effective_timeout_ms); |
| 746 |
1/2
✓ Branch 1 taken 12 times.
✗ Branch 2 not taken.
|
12 |
request.set_protocol_version(endpoint_.protocol_version); |
| 747 |
1/2
✓ Branch 1 taken 12 times.
✗ Branch 2 not taken.
|
12 |
request.set_deployment_id(endpoint_.deployment_id); |
| 748 |
1/2
✓ Branch 1 taken 12 times.
✗ Branch 2 not taken.
|
12 |
request.set_deployment_epoch(endpoint_.deployment_epoch); |
| 749 |
1/2
✓ Branch 1 taken 12 times.
✗ Branch 2 not taken.
|
12 |
request.set_configuration_digest(endpoint_.configuration_digest); |
| 750 |
1/2
✓ Branch 1 taken 12 times.
✗ Branch 2 not taken.
|
12 |
request.set_fabric_digest(endpoint_.fabric_digest); |
| 751 |
|
|
|
| 752 |
1/2
✓ Branch 1 taken 12 times.
✗ Branch 2 not taken.
|
12 |
WaitServerResponse response; |
| 753 |
1/2
✓ Branch 1 taken 12 times.
✗ Branch 2 not taken.
|
12 |
grpc::ClientContext context; |
| 754 |
2/4
✓ Branch 1 taken 12 times.
✗ Branch 2 not taken.
✓ Branch 4 taken 12 times.
✗ Branch 5 not taken.
|
12 |
context.set_deadline(DeadlineFromNow(effective_timeout_ms)); |
| 755 |
4/6
✓ Branch 2 taken 12 times.
✗ Branch 3 not taken.
✓ Branch 5 taken 12 times.
✗ Branch 6 not taken.
✓ Branch 8 taken 6 times.
✓ Branch 9 taken 6 times.
|
30 |
ThrowIfNotOk(stub->WaitServer(&context, request, &response), "wait_server"); |
| 756 |
|
24 |
} |
| 757 |
|
|
|
| 758 |
|
8 |
void RdmaControlPlaneClient::WaitServerReady(int num_servers, |
| 759 |
|
|
int timeout_ms) const { |
| 760 |
|
8 |
const int effective_timeout_ms = |
| 761 |
1/2
✓ Branch 0 taken 8 times.
✗ Branch 1 not taken.
|
8 |
timeout_ms > 0 ? timeout_ms : endpoint_.timeout_ms; |
| 762 |
|
8 |
RdmaControlPlane::Stub* stub = channels_->stub.get(); |
| 763 |
|
|
|
| 764 |
1/2
✓ Branch 1 taken 8 times.
✗ Branch 2 not taken.
|
8 |
WaitServerReadyRequest request; |
| 765 |
1/2
✓ Branch 1 taken 8 times.
✗ Branch 2 not taken.
|
8 |
request.set_num_servers(num_servers); |
| 766 |
1/2
✓ Branch 1 taken 8 times.
✗ Branch 2 not taken.
|
8 |
request.set_timeout_ms(effective_timeout_ms); |
| 767 |
1/2
✓ Branch 1 taken 8 times.
✗ Branch 2 not taken.
|
8 |
request.set_protocol_version(endpoint_.protocol_version); |
| 768 |
1/2
✓ Branch 1 taken 8 times.
✗ Branch 2 not taken.
|
8 |
request.set_deployment_id(endpoint_.deployment_id); |
| 769 |
1/2
✓ Branch 1 taken 8 times.
✗ Branch 2 not taken.
|
8 |
request.set_deployment_epoch(endpoint_.deployment_epoch); |
| 770 |
1/2
✓ Branch 1 taken 8 times.
✗ Branch 2 not taken.
|
8 |
request.set_configuration_digest(endpoint_.configuration_digest); |
| 771 |
1/2
✓ Branch 1 taken 8 times.
✗ Branch 2 not taken.
|
8 |
request.set_fabric_digest(endpoint_.fabric_digest); |
| 772 |
|
|
|
| 773 |
1/2
✓ Branch 1 taken 8 times.
✗ Branch 2 not taken.
|
8 |
WaitServerReadyResponse response; |
| 774 |
1/2
✓ Branch 1 taken 8 times.
✗ Branch 2 not taken.
|
8 |
grpc::ClientContext context; |
| 775 |
2/4
✓ Branch 1 taken 8 times.
✗ Branch 2 not taken.
✓ Branch 4 taken 8 times.
✗ Branch 5 not taken.
|
8 |
context.set_deadline(DeadlineFromNow(effective_timeout_ms)); |
| 776 |
4/6
✓ Branch 2 taken 8 times.
✗ Branch 3 not taken.
✓ Branch 5 taken 8 times.
✗ Branch 6 not taken.
✓ Branch 8 taken 2 times.
✓ Branch 9 taken 6 times.
|
26 |
ThrowIfNotOk(stub->WaitServerReady(&context, request, &response), |
| 777 |
|
|
"wait_server_ready"); |
| 778 |
|
20 |
} |
| 779 |
|
|
|
| 780 |
|
26 |
RdmaControlPlaneServer::RdmaControlPlaneServer( |
| 781 |
|
26 |
RdmaControlPlaneEndpoint endpoint) |
| 782 |
|
26 |
: endpoint_(std::move(endpoint)) {} |
| 783 |
|
|
|
| 784 |
|
26 |
RdmaControlPlaneServer::~RdmaControlPlaneServer() { Stop(); } |
| 785 |
|
|
|
| 786 |
|
|
std::size_t |
| 787 |
|
8 |
RdmaControlPlaneServer::MetaKeyHash::operator()(const MetaKey& key) const { |
| 788 |
|
8 |
std::size_t hash = std::hash<std::string>{}(key.deployment_id); |
| 789 |
|
8 |
hash = hash * 1315423911u + static_cast<std::size_t>(key.deployment_epoch); |
| 790 |
|
8 |
hash = hash * 1315423911u + static_cast<std::size_t>(key.publisher_node_id); |
| 791 |
|
8 |
hash = hash * 1315423911u + static_cast<std::size_t>(key.publisher_lane); |
| 792 |
|
8 |
hash = hash * 1315423911u + static_cast<std::size_t>(key.receiver_node_id); |
| 793 |
|
8 |
hash = hash * 1315423911u + static_cast<std::size_t>(key.receiver_lane); |
| 794 |
|
8 |
return hash; |
| 795 |
|
|
} |
| 796 |
|
|
|
| 797 |
|
|
std::size_t |
| 798 |
|
98 |
RdmaControlPlaneServer::ReadyKeyHash::operator()(const ReadyKey& key) const { |
| 799 |
|
98 |
std::size_t hash = std::hash<std::string>{}(key.deployment_id); |
| 800 |
|
98 |
hash = hash * 1315423911u + static_cast<std::size_t>(key.deployment_epoch); |
| 801 |
|
98 |
hash = hash * 1315423911u + static_cast<std::size_t>(key.server_id); |
| 802 |
|
98 |
return hash; |
| 803 |
|
|
} |
| 804 |
|
|
|
| 805 |
|
12 |
int RdmaControlPlaneServer::ReadyCount( |
| 806 |
|
|
const std::string& deployment_id, |
| 807 |
|
|
std::uint64_t deployment_epoch, |
| 808 |
|
|
std::uint32_t protocol_version, |
| 809 |
|
|
const std::string& configuration_digest, |
| 810 |
|
|
const std::string& fabric_digest) const { |
| 811 |
|
12 |
int count = 0; |
| 812 |
2/2
✓ Branch 5 taken 8 times.
✓ Branch 6 taken 12 times.
|
20 |
for (const auto& entry : ready_servers_) { |
| 813 |
|
8 |
if (entry.first.deployment_id == deployment_id && |
| 814 |
1/2
✓ Branch 0 taken 8 times.
✗ Branch 1 not taken.
|
8 |
entry.first.deployment_epoch == deployment_epoch && |
| 815 |
2/2
✓ Branch 0 taken 6 times.
✓ Branch 1 taken 2 times.
|
8 |
entry.second.state == ReadyRecord::State::kServing && |
| 816 |
2/4
✓ Branch 0 taken 6 times.
✗ Branch 1 not taken.
✓ Branch 2 taken 6 times.
✗ Branch 3 not taken.
|
12 |
entry.second.protocol_version == protocol_version && |
| 817 |
4/6
✓ Branch 0 taken 8 times.
✗ Branch 1 not taken.
✓ Branch 3 taken 6 times.
✗ Branch 4 not taken.
✓ Branch 5 taken 6 times.
✓ Branch 6 taken 2 times.
|
22 |
entry.second.configuration_digest == configuration_digest && |
| 818 |
|
6 |
entry.second.fabric_digest == fabric_digest) { |
| 819 |
|
6 |
++count; |
| 820 |
|
|
} |
| 821 |
|
|
} |
| 822 |
|
12 |
return count; |
| 823 |
|
|
} |
| 824 |
|
|
|
| 825 |
|
14 |
bool RdmaControlPlaneServer::HasReadyContractMismatch( |
| 826 |
|
|
const std::string& deployment_id, |
| 827 |
|
|
std::uint64_t deployment_epoch, |
| 828 |
|
|
std::uint32_t protocol_version, |
| 829 |
|
|
const std::string& configuration_digest, |
| 830 |
|
|
const std::string& fabric_digest) const { |
| 831 |
2/2
✓ Branch 5 taken 10 times.
✓ Branch 6 taken 14 times.
|
24 |
for (const auto& entry : ready_servers_) { |
| 832 |
|
10 |
if (entry.first.deployment_id == deployment_id && |
| 833 |
3/6
✓ Branch 0 taken 10 times.
✗ Branch 1 not taken.
✓ Branch 2 taken 10 times.
✗ Branch 3 not taken.
✗ Branch 4 not taken.
✓ Branch 5 taken 10 times.
|
20 |
entry.first.deployment_epoch == deployment_epoch && |
| 834 |
2/4
✓ Branch 0 taken 10 times.
✗ Branch 1 not taken.
✓ Branch 2 taken 10 times.
✗ Branch 3 not taken.
|
20 |
(entry.second.protocol_version != protocol_version || |
| 835 |
1/2
✗ Branch 1 not taken.
✓ Branch 2 taken 10 times.
|
20 |
entry.second.configuration_digest != configuration_digest || |
| 836 |
|
10 |
entry.second.fabric_digest != fabric_digest)) { |
| 837 |
|
✗ |
return true; |
| 838 |
|
|
} |
| 839 |
|
|
} |
| 840 |
|
14 |
return false; |
| 841 |
|
|
} |
| 842 |
|
|
|
| 843 |
|
14 |
bool RdmaControlPlaneServer::HasDrainingServer( |
| 844 |
|
|
const std::string& deployment_id, |
| 845 |
|
|
std::uint64_t deployment_epoch, |
| 846 |
|
|
int num_servers) const { |
| 847 |
2/2
✓ Branch 5 taken 10 times.
✓ Branch 6 taken 10 times.
|
20 |
for (const auto& entry : ready_servers_) { |
| 848 |
|
10 |
if (entry.first.deployment_id == deployment_id && |
| 849 |
1/2
✓ Branch 0 taken 10 times.
✗ Branch 1 not taken.
|
10 |
entry.first.deployment_epoch == deployment_epoch && |
| 850 |
5/8
✓ Branch 0 taken 10 times.
✗ Branch 1 not taken.
✓ Branch 2 taken 10 times.
✗ Branch 3 not taken.
✓ Branch 4 taken 10 times.
✗ Branch 5 not taken.
✓ Branch 6 taken 4 times.
✓ Branch 7 taken 6 times.
|
30 |
entry.first.server_id >= 0 && entry.first.server_id < num_servers && |
| 851 |
2/2
✓ Branch 0 taken 4 times.
✓ Branch 1 taken 6 times.
|
10 |
entry.second.state == ReadyRecord::State::kDraining) { |
| 852 |
|
4 |
return true; |
| 853 |
|
|
} |
| 854 |
|
|
} |
| 855 |
|
10 |
return false; |
| 856 |
|
|
} |
| 857 |
|
|
|
| 858 |
|
|
RdmaServerHealthSnapshot |
| 859 |
|
36 |
RdmaControlPlaneServer::ServerHealthSnapshot(int server_id) const { |
| 860 |
1/2
✓ Branch 1 taken 36 times.
✗ Branch 2 not taken.
|
36 |
std::lock_guard<std::mutex> guard(mu_); |
| 861 |
|
72 |
const auto it = ready_servers_.find( |
| 862 |
2/4
✓ Branch 1 taken 36 times.
✗ Branch 2 not taken.
✓ Branch 4 taken 36 times.
✗ Branch 5 not taken.
|
36 |
ReadyKey{endpoint_.deployment_id, endpoint_.deployment_epoch, server_id}); |
| 863 |
2/2
✓ Branch 2 taken 4 times.
✓ Branch 3 taken 32 times.
|
36 |
if (it == ready_servers_.end()) { |
| 864 |
|
4 |
return RdmaServerHealthSnapshot{}; |
| 865 |
|
|
} |
| 866 |
|
32 |
const auto now = std::chrono::steady_clock::now(); |
| 867 |
|
32 |
RdmaServerHealthSnapshot snapshot; |
| 868 |
|
32 |
snapshot.known = true; |
| 869 |
1/2
✓ Branch 1 taken 32 times.
✗ Branch 2 not taken.
|
32 |
snapshot.health = EvaluateServerHealth( |
| 870 |
|
32 |
it->second.last_seen, |
| 871 |
|
|
now, |
| 872 |
|
32 |
endpoint_.health_stale_ms, |
| 873 |
|
32 |
endpoint_.health_dead_ms); |
| 874 |
2/4
✓ Branch 3 taken 32 times.
✗ Branch 4 not taken.
✓ Branch 5 taken 32 times.
✗ Branch 6 not taken.
|
32 |
if (it->second.last_seen != std::chrono::steady_clock::time_point{}) { |
| 875 |
|
32 |
snapshot.last_seen_age_ms = |
| 876 |
1/2
✓ Branch 1 taken 32 times.
✗ Branch 2 not taken.
|
32 |
std::chrono::duration_cast<std::chrono::milliseconds>( |
| 877 |
1/2
✓ Branch 2 taken 32 times.
✗ Branch 3 not taken.
|
64 |
now - it->second.last_seen) |
| 878 |
|
32 |
.count(); |
| 879 |
|
|
} |
| 880 |
|
32 |
return snapshot; |
| 881 |
|
36 |
} |
| 882 |
|
|
|
| 883 |
|
18 |
void RdmaControlPlaneServer::RecordHeartbeat(int server_id) { |
| 884 |
1/2
✓ Branch 1 taken 18 times.
✗ Branch 2 not taken.
|
18 |
std::lock_guard<std::mutex> guard(mu_); |
| 885 |
|
36 |
const auto it = ready_servers_.find( |
| 886 |
2/4
✓ Branch 1 taken 18 times.
✗ Branch 2 not taken.
✓ Branch 4 taken 18 times.
✗ Branch 5 not taken.
|
18 |
ReadyKey{endpoint_.deployment_id, endpoint_.deployment_epoch, server_id}); |
| 887 |
2/2
✓ Branch 2 taken 4 times.
✓ Branch 3 taken 14 times.
|
18 |
if (it == ready_servers_.end()) { |
| 888 |
|
|
// Unknown shard: not ready yet, or never ready. A heartbeat never creates a |
| 889 |
|
|
// record; READY remains the only way to become serving. |
| 890 |
|
4 |
return; |
| 891 |
|
|
} |
| 892 |
|
14 |
it->second.last_seen = std::chrono::steady_clock::now(); |
| 893 |
2/2
✓ Branch 1 taken 14 times.
✓ Branch 2 taken 4 times.
|
18 |
} |
| 894 |
|
|
|
| 895 |
|
26 |
void RdmaControlPlaneServer::Start() { |
| 896 |
1/2
✗ Branch 1 not taken.
✓ Branch 2 taken 26 times.
|
26 |
if (server_ != nullptr) { |
| 897 |
|
✗ |
return; |
| 898 |
|
|
} |
| 899 |
2/4
✓ Branch 1 taken 26 times.
✗ Branch 2 not taken.
✓ Branch 3 taken 26 times.
✗ Branch 4 not taken.
|
52 |
if (endpoint_.host.empty() || endpoint_.port < 1 || endpoint_.port > 65535 || |
| 900 |
2/4
✓ Branch 0 taken 26 times.
✗ Branch 1 not taken.
✓ Branch 3 taken 26 times.
✗ Branch 4 not taken.
|
26 |
endpoint_.timeout_ms <= 0 || endpoint_.deployment_id.empty() || |
| 901 |
1/2
✓ Branch 0 taken 26 times.
✗ Branch 1 not taken.
|
26 |
endpoint_.deployment_epoch == 0 || |
| 902 |
2/4
✓ Branch 0 taken 26 times.
✗ Branch 1 not taken.
✓ Branch 2 taken 26 times.
✗ Branch 3 not taken.
|
52 |
endpoint_.protocol_version != recstore::kRdmaDeploymentProtocolVersion || |
| 903 |
3/6
✓ Branch 0 taken 26 times.
✗ Branch 1 not taken.
✗ Branch 3 not taken.
✓ Branch 4 taken 26 times.
✗ Branch 5 not taken.
✓ Branch 6 taken 26 times.
|
78 |
endpoint_.configuration_digest.empty() || |
| 904 |
|
26 |
endpoint_.fabric_digest.empty()) { |
| 905 |
0/2
✗ Branch 2 not taken.
✗ Branch 3 not taken.
|
✗ |
throw std::invalid_argument("invalid RDMA control-plane endpoint contract"); |
| 906 |
|
|
} |
| 907 |
|
26 |
stop_requested_.store(false, std::memory_order_relaxed); |
| 908 |
1/2
✓ Branch 1 taken 26 times.
✗ Branch 2 not taken.
|
26 |
service_ = std::make_unique<RdmaControlPlaneService>(this); |
| 909 |
1/2
✓ Branch 1 taken 26 times.
✗ Branch 2 not taken.
|
26 |
grpc::ServerBuilder builder; |
| 910 |
1/2
✓ Branch 1 taken 26 times.
✗ Branch 2 not taken.
|
26 |
builder.AddListeningPort( |
| 911 |
2/4
✓ Branch 1 taken 26 times.
✗ Branch 2 not taken.
✓ Branch 4 taken 26 times.
✗ Branch 5 not taken.
|
52 |
EndpointString(endpoint_), grpc::InsecureServerCredentials()); |
| 912 |
1/2
✓ Branch 2 taken 26 times.
✗ Branch 3 not taken.
|
26 |
builder.RegisterService(service_.get()); |
| 913 |
1/2
✓ Branch 1 taken 26 times.
✗ Branch 2 not taken.
|
26 |
server_ = builder.BuildAndStart(); |
| 914 |
1/2
✗ Branch 1 not taken.
✓ Branch 2 taken 26 times.
|
26 |
if (server_ == nullptr) { |
| 915 |
0/4
✗ Branch 1 not taken.
✗ Branch 2 not taken.
✗ Branch 4 not taken.
✗ Branch 5 not taken.
|
✗ |
throw std::runtime_error("control-plane gRPC server failed to listen on " + |
| 916 |
0/2
✗ Branch 2 not taken.
✗ Branch 3 not taken.
|
✗ |
EndpointString(endpoint_)); |
| 917 |
|
|
} |
| 918 |
|
26 |
} |
| 919 |
|
|
|
| 920 |
|
52 |
void RdmaControlPlaneServer::Stop() { |
| 921 |
2/2
✓ Branch 1 taken 26 times.
✓ Branch 2 taken 26 times.
|
52 |
if (server_ == nullptr) { |
| 922 |
|
26 |
return; |
| 923 |
|
|
} |
| 924 |
|
26 |
stop_requested_.store(true, std::memory_order_relaxed); |
| 925 |
|
|
{ |
| 926 |
1/2
✓ Branch 1 taken 26 times.
✗ Branch 2 not taken.
|
26 |
std::lock_guard<std::mutex> guard(mu_); |
| 927 |
|
26 |
cv_.notify_all(); |
| 928 |
|
26 |
} |
| 929 |
|
26 |
server_->Shutdown(); |
| 930 |
|
26 |
server_.reset(); |
| 931 |
|
26 |
service_.reset(); |
| 932 |
|
|
} |
| 933 |
|
|
|
| 934 |
|
|
} // namespace petps |
| 935 |
|
|
|