Skip to content

Commit 1d024d9

Browse files
authored
Add an authenticated direct-query PEM mode for vizier deployments (#2408)
Summary: Add an authenticated direct-query PEM mode for vizier deployments Exposes VizierService.ExecuteScript directly on the normal PEM over a dedicated port (:50305 default), so node-local clients can query Carnot without a broker hop or cloud dependency. Relevant Issues: N/A Type of change: /kind feature Test Plan: New tests verify new functionality. k8sstormcenter project has been using an earlier version of this change (#2401) end to end in vizier deployments Changelog Message: Add a direct-query PEM mode (disabled by default) to vizier deployments --------- Signed-off-by: Dom Del Nano <ddelnano@gmail.com>
1 parent 50cabe7 commit 1d024d9

13 files changed

Lines changed: 721 additions & 22 deletions

‎src/carnot/blocking_agg_benchmark.cc‎

Lines changed: 4 additions & 4 deletions
Original file line numberDiff line numberDiff line change
@@ -89,9 +89,9 @@ void BM_Query(benchmark::State& state, std::vector<types::DataType> types,
8989
int64_t num_batches, const datagen::DistributionParams* dist_vars,
9090
const datagen::DistributionParams* len_vars) {
9191
auto table_store = std::make_shared<table_store::TableStore>();
92-
auto server = LocalGRPCResultSinkServer();
92+
auto server = LocalGRPCResultSinkServer::Create().ConsumeValueOrDie();
9393

94-
auto carnot = SetUpCarnot(table_store, &server);
94+
auto carnot = SetUpCarnot(table_store, server.get());
9595
auto table = table_store::CreateTable(types, distribution_types, state.range(0), num_batches,
9696
dist_vars, len_vars)
9797
.ConsumeValueOrDie();
@@ -105,8 +105,8 @@ void BM_Query(benchmark::State& state, std::vector<types::DataType> types,
105105
if (!res.ok()) {
106106
LOG(FATAL) << "Aggregate benchmark query did not execute successfully.";
107107
}
108-
bytes_processed += server.exec_stats().ConsumeValueOrDie().execution_stats().bytes_processed();
109-
server.ResetQueryResults();
108+
bytes_processed += server->exec_stats().ConsumeValueOrDie().execution_stats().bytes_processed();
109+
server->ResetQueryResults();
110110
++i;
111111
}
112112

‎src/carnot/carnot_executable.cc‎

Lines changed: 10 additions & 5 deletions
Original file line numberDiff line numberDiff line change
@@ -240,13 +240,18 @@ int main(int argc, char* argv[]) {
240240

241241
// Execute query.
242242
auto table_store = std::make_shared<px::table_store::TableStore>();
243-
auto result_server = px::carnot::exec::LocalGRPCResultSinkServer();
243+
auto result_server_or = px::carnot::exec::LocalGRPCResultSinkServer::Create();
244+
if (!result_server_or.ok()) {
245+
LOG(FATAL) << absl::Substitute("Failed to start result server: $0",
246+
result_server_or.status().msg());
247+
}
248+
auto result_server = result_server_or.ConsumeValueOrDie();
244249
auto func_registry = std::make_unique<px::carnot::udf::Registry>("default_registry");
245250
px::carnot::funcs::RegisterFuncsOrDie(func_registry.get());
246251
auto clients_config =
247252
std::make_unique<px::carnot::Carnot::ClientsConfig>(px::carnot::Carnot::ClientsConfig{
248253
[&result_server](const std::string& address, const std::string&) {
249-
return result_server.StubGenerator(address);
254+
return result_server->StubGenerator(address);
250255
},
251256
[](grpc::ClientContext*) {},
252257
});
@@ -263,13 +268,13 @@ int main(int argc, char* argv[]) {
263268
LOG(FATAL) << absl::Substitute("Query failed to execute: $0", exec_status.msg());
264269
}
265270

266-
auto output_names = result_server.output_tables();
271+
auto output_names = result_server->output_tables();
267272
if (!output_names.size()) {
268273
LOG(FATAL) << "Query produced no output tables.";
269274
}
270-
std::string output_name = *(result_server.output_tables().begin());
275+
std::string output_name = *(result_server->output_tables().begin());
271276
LOG(INFO) << absl::Substitute("Writing results for output table: $0", output_name);
272277
// Write output table to CSV.
273-
TableToCsv(output_filename, result_server.query_results(output_name));
278+
TableToCsv(output_filename, result_server->query_results(output_name));
274279
return 0;
275280
}

‎src/carnot/carnot_test.cc‎

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -50,7 +50,7 @@ class CarnotTest : public ::testing::Test {
5050
void SetUp() override {
5151
Test::SetUp();
5252
table_store_ = std::make_shared<table_store::TableStore>();
53-
result_server_ = std::make_unique<exec::LocalGRPCResultSinkServer>();
53+
ASSERT_OK_AND_ASSIGN(result_server_, exec::LocalGRPCResultSinkServer::Create());
5454

5555
auto func_registry = std::make_unique<px::carnot::udf::Registry>("default_registry");
5656
funcs::RegisterFuncsOrDie(func_registry.get());

‎src/carnot/end_to_end_join_test.cc‎

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -44,7 +44,7 @@ class JoinTest : public ::testing::Test {
4444
void SetUp() override {
4545
Test::SetUp();
4646
table_store_ = std::make_shared<table_store::TableStore>();
47-
result_server_ = std::make_unique<exec::LocalGRPCResultSinkServer>();
47+
ASSERT_OK_AND_ASSIGN(result_server_, exec::LocalGRPCResultSinkServer::Create());
4848
auto func_registry = std::make_unique<px::carnot::udf::Registry>("default_registry");
4949
px::carnot::funcs::RegisterFuncsOrDie(func_registry.get());
5050
auto clients_config = std::make_unique<Carnot::ClientsConfig>(Carnot::ClientsConfig{

‎src/carnot/engine_state.h‎

Lines changed: 6 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -128,14 +128,18 @@ class EngineState : public NotCopyable {
128128
return std::make_unique<plan::PlanState>(func_registry_.get());
129129
}
130130

131+
// max_output_rows_per_table of 0 means unlimited. That is the right default
132+
// for the single-node executable and for tests, where the caller controls the
133+
// input, but a server accepting queries from a peer should pass a real cap:
134+
// results are buffered in memory before they are streamed out.
131135
std::unique_ptr<planner::CompilerState> CreateLocalExecutionCompilerState(
132-
types::Time64NSValue time_now) {
136+
types::Time64NSValue time_now, int64_t max_output_rows_per_table = 0) {
133137
auto rel_map = table_store_->GetRelationMap();
134138
// Use an empty string for query result address, because the local execution mode should use
135139
// the Local GRPC result server to send results to.
136140
return std::make_unique<planner::CompilerState>(
137141
std::move(rel_map), planner::SensitiveColumnMap{}, registry_info_.get(), time_now,
138-
/* max_output_rows_per_table */ 0,
142+
max_output_rows_per_table,
139143
/* result address */ "",
140144
/* ssl target name override*/ "", planner::RedactionOptions{}, nullptr, nullptr,
141145
planner::DebugInfo{});

‎src/carnot/exec/local_grpc_result_server.h‎

Lines changed: 22 additions & 8 deletions
Original file line numberDiff line numberDiff line change
@@ -64,7 +64,12 @@ class LocalResultSinkServer final : public carnotpb::ResultSinkService::Service
6464
return ::grpc::Status::OK;
6565
}
6666

67-
void ResetQueryResults() { query_results_.clear(); }
67+
void ResetQueryResults() {
68+
// Takes the same lock TransferResultChunk does: a caller resetting between
69+
// queries races an in-flight chunk write otherwise.
70+
const std::lock_guard<std::mutex> lock(result_mutex_);
71+
query_results_.clear();
72+
}
6873

6974
private:
7075
// List of the query results received.
@@ -75,13 +80,14 @@ class LocalResultSinkServer final : public carnotpb::ResultSinkService::Service
7580

7681
class LocalGRPCResultSinkServer {
7782
public:
78-
LocalGRPCResultSinkServer() {
79-
grpc::ServerBuilder builder;
80-
81-
builder.AddListeningPort("127.0.0.1:0", grpc::InsecureServerCredentials());
82-
builder.RegisterService(&result_sink_server_);
83-
grpc_server_ = builder.BuildAndStart();
84-
CHECK(grpc_server_ != nullptr);
83+
// Starts the in-process result sink server. Returns an error if it fails to bind.
84+
static StatusOr<std::unique_ptr<LocalGRPCResultSinkServer>> Create() {
85+
auto server = std::unique_ptr<LocalGRPCResultSinkServer>(new LocalGRPCResultSinkServer());
86+
if (server->grpc_server_ == nullptr) {
87+
return error::ResourceUnavailable(
88+
"failed to start the local result sink server on 127.0.0.1:0");
89+
}
90+
return server;
8591
}
8692

8793
~LocalGRPCResultSinkServer() {
@@ -151,6 +157,14 @@ class LocalGRPCResultSinkServer {
151157
}
152158

153159
private:
160+
LocalGRPCResultSinkServer() {
161+
grpc::ServerBuilder builder;
162+
163+
builder.AddListeningPort("127.0.0.1:0", grpc::InsecureServerCredentials());
164+
builder.RegisterService(&result_sink_server_);
165+
grpc_server_ = builder.BuildAndStart();
166+
}
167+
154168
std::unique_ptr<grpc::Server> grpc_server_;
155169
LocalResultSinkServer result_sink_server_;
156170
};

‎src/vizier/services/agent/pem/BUILD.bazel‎

Lines changed: 26 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -31,12 +31,22 @@ pl_cc_library(
3131
),
3232
hdrs = glob(["*.h"]),
3333
deps = [
34+
"//src/api/proto/vizierpb:vizier_pl_cc_proto",
35+
"//src/carnot",
36+
"//src/carnot/carnotpb:carnot_pl_cc_proto",
37+
"//src/carnot/funcs:cc_library",
38+
"//src/carnot/planner/compiler:cc_library",
3439
"//src/carnot/planner/dynamic_tracing/ir/logicalpb:logical_pl_cc_proto",
40+
"//src/carnot/planpb:plan_pl_cc_proto",
3541
"//src/integrations/grpc_clocksync:cc_library",
42+
"//src/shared/services/jwt:cc_library",
3643
"//src/shared/tracepoint_translation:cc_library",
3744
"//src/stirling:cc_library",
3845
"//src/stirling/source_connectors/dynamic_tracer/dynamic_tracing/ir/logicalpb:logical_pl_cc_proto",
3946
"//src/vizier/services/agent/shared/manager:cc_library",
47+
"//src/vizier/services/agent/shared/vizier_results:cc_library",
48+
"@com_github_grpc_grpc//:grpc++",
49+
"@com_github_rlyeh_sole//:sole",
4050
],
4151
)
4252

@@ -51,6 +61,22 @@ pl_cc_test(
5161
],
5262
)
5363

64+
pl_cc_test(
65+
name = "direct_query_server_test",
66+
srcs = ["direct_query_server_test.cc"],
67+
deps = [
68+
":cc_library",
69+
"//src/api/proto/vizierpb:vizier_pl_cc_proto",
70+
"//src/carnot",
71+
"//src/carnot/funcs:cc_library",
72+
"//src/shared/services/jwt:cc_library",
73+
"//src/shared/types:cc_library",
74+
"//src/table_store:cc_library",
75+
"@com_github_grpc_grpc//:grpc++",
76+
"@com_github_rlyeh_sole//:sole",
77+
],
78+
)
79+
5480
pl_cc_binary(
5581
name = "pem",
5682
srcs = ["pem_main.cc"],
Lines changed: 165 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,165 @@
1+
/*
2+
* Copyright 2018- The Pixie Authors.
3+
*
4+
* Licensed under the Apache License, Version 2.0 (the "License");
5+
* you may not use this file except in compliance with the License.
6+
* You may obtain a copy of the License at
7+
*
8+
* http://www.apache.org/licenses/LICENSE-2.0
9+
*
10+
* Unless required by applicable law or agreed to in writing, software
11+
* distributed under the License is distributed on an "AS IS" BASIS,
12+
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13+
* See the License for the specific language governing permissions and
14+
* limitations under the License.
15+
*
16+
* SPDX-License-Identifier: Apache-2.0
17+
*/
18+
19+
#include "src/vizier/services/agent/pem/direct_query_server.h"
20+
21+
#include <chrono>
22+
#include <cstdint>
23+
#include <cstring>
24+
#include <iterator>
25+
#include <string>
26+
#include <string_view>
27+
#include <utility>
28+
#include <vector>
29+
30+
#include <absl/strings/str_split.h>
31+
#include <absl/strings/string_view.h>
32+
#include <absl/strings/substitute.h>
33+
#include <jwt/base64.hpp>
34+
#include <sole.hpp>
35+
36+
#include "src/carnot/carnot.h"
37+
#include "src/carnot/carnotpb/carnot.pb.h"
38+
#include "src/carnot/engine_state.h"
39+
#include "src/carnot/exec/local_grpc_result_server.h"
40+
#include "src/carnot/planner/compiler/compiler.h"
41+
#include "src/carnot/planpb/plan.pb.h"
42+
#include "src/common/base/base.h"
43+
#include "src/shared/services/jwt/service_token.h"
44+
#include "src/shared/types/typespb/wrapper/types_pb_wrapper.h"
45+
#include "src/vizier/services/agent/shared/vizier_results/result_conversion.h"
46+
47+
namespace px {
48+
namespace vizier {
49+
namespace agent {
50+
51+
namespace {
52+
53+
constexpr char kAuthorizationMetadataKey[] = "authorization";
54+
55+
} // namespace
56+
57+
::grpc::Status AuthenticateRequest(::grpc::ServerContext* ctx, const std::string& jwt_signing_key) {
58+
if (jwt_signing_key.empty()) {
59+
return ::grpc::Status(::grpc::StatusCode::UNAUTHENTICATED,
60+
"direct-query: signing key not configured");
61+
}
62+
const auto& md = ctx->client_metadata();
63+
const auto range = md.equal_range(kAuthorizationMetadataKey);
64+
if (range.first == range.second) {
65+
return ::grpc::Status(::grpc::StatusCode::UNAUTHENTICATED,
66+
"direct-query: missing authorization metadata");
67+
}
68+
if (std::next(range.first) != range.second) {
69+
return ::grpc::Status(::grpc::StatusCode::UNAUTHENTICATED,
70+
"direct-query: duplicate authorization metadata");
71+
}
72+
const std::string_view raw(range.first->second.data(), range.first->second.size());
73+
const std::string_view token = ::px::services::StripBearerPrefix(raw);
74+
if (token.empty()) {
75+
return ::grpc::Status(::grpc::StatusCode::UNAUTHENTICATED,
76+
"direct-query: authorization is not a Bearer token");
77+
}
78+
const auto s = ::px::services::VerifyServiceJWT(token, jwt_signing_key);
79+
if (!s.ok()) {
80+
VLOG(1) << "direct-query: rejecting bearer token: " << s.msg();
81+
return ::grpc::Status(::grpc::StatusCode::UNAUTHENTICATED,
82+
"direct-query: invalid bearer token");
83+
}
84+
return ::grpc::Status::OK;
85+
}
86+
87+
namespace {
88+
89+
constexpr int64_t kMaxOutputRowsPerTable = 10000;
90+
91+
std::vector<::px::api::vizierpb::ExecuteScriptResponse> collectResponses(
92+
::px::carnot::exec::LocalGRPCResultSinkServer* result_server, const std::string& query_id) {
93+
std::vector<::px::api::vizierpb::ExecuteScriptResponse> out;
94+
for (const auto& chunk : result_server->raw_query_results()) {
95+
auto resp = ExecuteScriptResponseFromChunk(chunk, query_id);
96+
if (!resp.has_value()) {
97+
continue;
98+
}
99+
out.push_back(std::move(*resp));
100+
}
101+
return out;
102+
}
103+
104+
} // namespace
105+
106+
::grpc::Status DirectQueryServer::ExecuteScript(
107+
::grpc::ServerContext* context, const ::px::api::vizierpb::ExecuteScriptRequest* request,
108+
::grpc::ServerWriter<::px::api::vizierpb::ExecuteScriptResponse>* writer) {
109+
if (auto s = AuthenticateRequest(context, jwt_signing_key_); !s.ok()) {
110+
return s;
111+
}
112+
if (request->mutation()) {
113+
return ::grpc::Status(::grpc::StatusCode::UNIMPLEMENTED,
114+
"direct-query: mutations are out of scope (read-only endpoint)");
115+
}
116+
if (carnot_ == nullptr || engine_state_ == nullptr || result_server_ == nullptr) {
117+
return ::grpc::Status(::grpc::StatusCode::FAILED_PRECONDITION,
118+
"direct-query: server not wired with a live Carnot");
119+
}
120+
const auto query_id = sole::uuid4();
121+
const std::string query_id_str = query_id.str();
122+
123+
const auto time_now = ::px::CurrentTimeNS();
124+
125+
auto compiler_state =
126+
engine_state_->CreateLocalExecutionCompilerState(time_now, kMaxOutputRowsPerTable);
127+
auto plan_or = ::px::carnot::planner::compiler::Compiler().Compile(request->query_str(),
128+
compiler_state.get());
129+
if (!plan_or.ok()) {
130+
auto msg = absl::Substitute("direct-query: PxL compile failed ($0)", plan_or.msg());
131+
VLOG(1) << msg;
132+
return ::grpc::Status(::grpc::StatusCode::INVALID_ARGUMENT, msg);
133+
}
134+
const auto plan = plan_or.ConsumeValueOrDie();
135+
EmitSchemaResponses(plan, query_id_str, writer);
136+
137+
std::vector<::px::api::vizierpb::ExecuteScriptResponse> responses;
138+
{
139+
absl::MutexLock lk(&exec_mu_);
140+
result_server_->ResetQueryResults();
141+
auto exec_s = carnot_->ExecuteQuery(request->query_str(), query_id, time_now);
142+
if (!exec_s.ok()) {
143+
auto msg = absl::Substitute("direct-query: PxL execute failed ($0)", exec_s.msg());
144+
VLOG(1) << msg;
145+
return ::grpc::Status(::grpc::StatusCode::INTERNAL, msg);
146+
}
147+
responses = collectResponses(result_server_, query_id_str);
148+
}
149+
150+
for (const auto& resp : responses) {
151+
if (context->IsCancelled()) {
152+
return ::grpc::Status(::grpc::StatusCode::CANCELLED, "direct-query: client cancelled");
153+
}
154+
if (!writer->Write(resp)) {
155+
VLOG(1) << "direct-query: client stream closed mid-response, abandoning query "
156+
<< query_id_str;
157+
break;
158+
}
159+
}
160+
return ::grpc::Status::OK;
161+
}
162+
163+
} // namespace agent
164+
} // namespace vizier
165+
} // namespace px

0 commit comments

Comments
 (0)