Files
cairnobs/ingest/internal/grpcserver/server.go
T
jcoffey-dev 1de77b969f Build per-tenant ClickHouse write-routing for ingest (Tantivy still deferred)
ingest tags every record with a tenant_id Kafka header (built previously),
but nothing consumed it to actually route the write. This closes that for
ClickHouse: enterprise/cmd/enterprise-ingest (a second binary, mirroring
enterprise-api) reuses ingest/consumer's own flush loop unchanged, with
enterprise/internal/chwriter.Registry -- a per-tenant clickhousewriter.Writer
registry -- swapped in as the writer. A batch pulled from the single shared
Redpanda topic can mix records from many tenants, so WriteBatch groups by
TenantID and dispatches each group to its own tenant's connection, fail-
closed on an empty or unrecognized tenant_id.

ingest/consumer and ingest/clickhousewriter move out of internal/ (same
reason api/internal/* moved earlier this phase: enterprise/ can't import
anything under another module's internal/). Their New() constructors now
take small local Config structs instead of ingest/internal/config types,
so enterprise/ doesn't need that import either.

Building this surfaced a real bug: tenantprovision.ProvisionClickHouse
only granted SELECT on a tenant's ClickHouse user, correct for chrunner's
read-only use but not enough for chwriter reusing the same credential to
write -- every real per-tenant write would have failed closed with a
permission error. Fixed by widening the grant to SELECT, INSERT; no
cross-tenant boundary is crossed by also allowing INSERT within a
tenant's own database.

Helm gates enterprise-ingest's Deployment on the same
ingest.requireTenantCredential flag that already gates tag validation --
write-routing is meaningless without tagging already being required, so
they're one decision, not two. docker-compose.yml's version is a
disclosed, weaker approximation: it can't achieve Helm's genuine
-mode=server/-mode=consumer split, so with the enterprise profile active
both ingest and enterprise-ingest independently consume every message
via different consumer groups -- harmless duplication for local
verification only.

Not built: Tantivy's independent Redpanda consumer (search/src/consumer.rs)
still doesn't read the tenant_id header at all -- every record still lands
in the one shared index regardless of tenant. Not run: the live-ClickHouse-
gated tests (chwriter's cross-tenant routing test, tenantprovision's INSERT
regression test) -- no Docker/database access in this environment; they're
correct Go that has never executed, disclosed as such in docs/security/
threat-model.md and docs/phase-4-runbook.md §14.
2026-08-14 19:26:09 -07:00

196 lines
7.5 KiB
Go

// Package grpcserver implements the agent-facing side of ingest: an mTLS
// gRPC server accepting LogIngest.PushBatch calls. It assigns each record
// a stable record_id (see the proto field comment for why this has to
// happen exactly once, here, rather than in either downstream consumer)
// and otherwise forwards records unchanged onto Redpanda — normalization
// into the ClickHouse row shape happens later, on the consumer side.
//
// If a TenantResolver is configured, PushBatch also resolves which
// tenant the call's bearer credential belongs to and attaches it as a
// "tenant_id" Kafka message header on every record produced -- the first
// step of Phase 4's ingest tenant-awareness (see
// /docs/phase-4-runbook.md and CLAUDE.md's "ingest itself has no tenant
// concept" gap). Deliberately scoped no further than that for now:
// nothing downstream (this package's own consumer, or `search`'s
// separate Redpanda consumer) reads that header yet to route a record's
// write into a per-tenant ClickHouse database/Tantivy index -- every
// record still lands in the one shared destination either way, tenant_id
// header or not. That's real, disclosed, deferred follow-up work, not
// silently incomplete: attaching a verifiable tenant identity as early
// as possible (right where the credential is actually presented) is a
// self-contained, independently valuable step on its own, and it's what
// any later per-tenant write-routing work will consume.
package grpcserver
import (
"context"
"fmt"
"log/slog"
"net"
"strings"
"github.com/google/uuid"
"github.com/segmentio/kafka-go"
"google.golang.org/grpc"
"google.golang.org/grpc/codes"
"google.golang.org/grpc/credentials"
"google.golang.org/grpc/metadata"
"google.golang.org/grpc/status"
"google.golang.org/protobuf/proto"
"github.com/sentry/sentry/ingest/internal/config"
logsv1 "github.com/sentry/sentry/proto/sentry/logs/v1"
)
// TenantIDHeaderKey is the Kafka message header a resolved tenant ID is
// attached under. ingest/consumer.TenantIDHeaderKey names the identical
// literal on the read side -- duplicated rather than imported (this
// package is the agent-facing producer side; consumer is a different
// concern, and importing across them for one string constant isn't
// worth the coupling), so a change here must be mirrored there.
const TenantIDHeaderKey = "tenant_id"
type Server struct {
logsv1.UnimplementedLogIngestServer
logger *slog.Logger
grpcCfg config.GRPCConfig
tlsCfg config.TLSConfig
producer batchProducer
resolver TenantResolver
}
// batchProducer is the subset of *producer.Producer this package depends
// on, so tests can substitute a fake without touching Redpanda.
type batchProducer interface {
WriteBatch(ctx context.Context, msgs []kafka.Message) error
}
// TenantResolver validates an ingest credential (a bearer token
// presented via gRPC metadata, `authorization: Bearer <token>`) and
// resolves which tenant it belongs to. nil is a deliberate no-op: every
// record's Kafka message gets no tenant_id header at all, matching every
// ingest deployment's behavior before per-tenant ingest credentials
// existed. The real implementation
// (ingest/internal/tenantresolver.HTTPResolver) is a plain HTTP client
// calling enterprise-auth's /internal/authorize-ingest -- never an
// enterprise/ import, since this package is AGPL core (same "network
// boundary, not import boundary" shape api/authz.Authorizer uses).
type TenantResolver interface {
ResolveTenant(ctx context.Context, token string) (tenantID string, err error)
}
func New(logger *slog.Logger, grpcCfg config.GRPCConfig, tlsCfg config.TLSConfig, p batchProducer, resolver TenantResolver) *Server {
return &Server{logger: logger, grpcCfg: grpcCfg, tlsCfg: tlsCfg, producer: p, resolver: resolver}
}
// Run blocks serving gRPC until ctx is canceled, then gracefully stops.
func (s *Server) Run(ctx context.Context) error {
tlsConf, err := loadServerTLSConfig(s.tlsCfg)
if err != nil {
return fmt.Errorf("loading TLS config: %w", err)
}
lis, err := net.Listen("tcp", s.grpcCfg.ListenAddr)
if err != nil {
return fmt.Errorf("listening on %s: %w", s.grpcCfg.ListenAddr, err)
}
grpcSrv := grpc.NewServer(grpc.Creds(credentials.NewTLS(tlsConf)))
logsv1.RegisterLogIngestServer(grpcSrv, s)
s.logger.Info("gRPC server listening", "addr", s.grpcCfg.ListenAddr)
errCh := make(chan error, 1)
go func() { errCh <- grpcSrv.Serve(lis) }()
select {
case <-ctx.Done():
grpcSrv.GracefulStop()
return nil
case err := <-errCh:
return err
}
}
func (s *Server) PushBatch(ctx context.Context, req *logsv1.PushBatchRequest) (*logsv1.PushBatchResponse, error) {
if len(req.GetRecords()) == 0 {
return &logsv1.PushBatchResponse{Accepted: 0}, nil
}
// tenantID stays empty (no header attached below) unless a resolver
// is actually configured -- single-tenant deployments never present
// a bearer credential and never need to. Once a resolver IS
// configured, a missing/invalid credential fails the whole batch
// closed rather than falling back to "no tenant" -- exactly the
// same fail-closed shape enterprise/internal/chrunner.Registry.RunSQL
// uses on the read side, applied here at the point data enters the
// system.
var tenantID string
if s.resolver != nil {
token, ok := bearerTokenFromContext(ctx)
if !ok {
return nil, status.Error(codes.Unauthenticated, "missing bearer credential")
}
resolved, err := s.resolver.ResolveTenant(ctx, token)
if err != nil {
s.logger.Error("resolving ingest tenant", "batch_id", req.GetBatchId(), "error", err)
return nil, status.Error(codes.Unauthenticated, "invalid ingest credential")
}
tenantID = resolved
}
msgs := make([]kafka.Message, 0, len(req.GetRecords()))
for _, rec := range req.GetRecords() {
// Assigned here, once, before this record is produced to
// Redpanda: the ClickHouse-writer consumer and the Tantivy-
// indexer consumer (Phase 1) both read the same Redpanda
// messages and need to agree on the same ID for the same
// record. Overwrites anything the agent sent (it always sends
// empty, per the proto comment, but this is authoritative
// regardless).
rec.RecordId = uuid.NewString()
val, err := proto.Marshal(rec)
if err != nil {
return nil, status.Errorf(codes.InvalidArgument, "marshaling record: %v", err)
}
msg := kafka.Message{
Key: []byte(rec.GetHost()),
Value: val,
}
if tenantID != "" {
msg.Headers = []kafka.Header{{Key: TenantIDHeaderKey, Value: []byte(tenantID)}}
}
msgs = append(msgs, msg)
}
if err := s.producer.WriteBatch(ctx, msgs); err != nil {
s.logger.Error("failed to write batch to redpanda", "batch_id", req.GetBatchId(), "error", err)
return nil, status.Errorf(codes.Unavailable, "writing to transport: %v", err)
}
s.logger.Debug("batch produced to redpanda", "batch_id", req.GetBatchId(), "records", len(req.GetRecords()), "tenant_id", tenantID)
return &logsv1.PushBatchResponse{Accepted: uint32(len(req.GetRecords()))}, nil
}
// bearerTokenFromContext reads the same "authorization: Bearer <token>"
// gRPC metadata shape HTTP's Authorization header uses -- an agent sets
// this once per PushBatch call (see the agent's grpc.rs), not per
// record.
func bearerTokenFromContext(ctx context.Context) (string, bool) {
md, ok := metadata.FromIncomingContext(ctx)
if !ok {
return "", false
}
values := md.Get("authorization")
if len(values) == 0 {
return "", false
}
const prefix = "Bearer "
if !strings.HasPrefix(values[0], prefix) {
return "", false
}
return strings.TrimPrefix(values[0], prefix), true
}