chore: add aibridge data to telemetry (#20449)

- Adds a new table to keep track of which payloads have already been
reported since we only report for the last clock hour
- Adds a query to gather and aggregate all the data by
provider/model/client

Relates to https://github.com/coder/coder-telemetry-server/issues/27
This commit is contained in:
Dean Sheather
2025-10-28 03:16:41 +11:00
committed by GitHub
parent cadf1352b4
commit 5a3ceb38f0
20 changed files with 940 additions and 6 deletions
+10
View File
@@ -24,6 +24,12 @@ const (
// but we won't touch the `connection_logs` table.
maxAuditLogConnectionEventAge = 90 * 24 * time.Hour // 90 days
auditLogConnectionEventBatchSize = 1000
// Telemetry heartbeats are used to deduplicate events across replicas. We
// don't need to persist heartbeat rows for longer than 24 hours, as they
// are only used for deduplication across replicas. The time needs to be
// long enough to cover the maximum interval of a heartbeat event (currently
// 1 hour) plus some buffer.
maxTelemetryHeartbeatAge = 24 * time.Hour
)
// New creates a new periodically purging database instance.
@@ -71,6 +77,10 @@ func New(ctx context.Context, logger slog.Logger, db database.Store, clk quartz.
if err := tx.ExpirePrebuildsAPIKeys(ctx, dbtime.Time(start)); err != nil {
return xerrors.Errorf("failed to expire prebuilds user api keys: %w", err)
}
deleteOldTelemetryLocksBefore := start.Add(-maxTelemetryHeartbeatAge)
if err := tx.DeleteOldTelemetryLocks(ctx, deleteOldTelemetryLocksBefore); err != nil {
return xerrors.Errorf("failed to delete old telemetry locks: %w", err)
}
deleteOldAuditLogConnectionEventsBefore := start.Add(-maxAuditLogConnectionEventAge)
if err := tx.DeleteOldAuditLogConnectionEvents(ctx, database.DeleteOldAuditLogConnectionEventsParams{
+53
View File
@@ -704,3 +704,56 @@ func TestExpireOldAPIKeys(t *testing.T) {
// Out of an abundance of caution, we do not expire explicitly named prebuilds API keys.
assertKeyActive(namedPrebuildsAPIKey.ID)
}
func TestDeleteOldTelemetryHeartbeats(t *testing.T) {
t.Parallel()
ctx := testutil.Context(t, testutil.WaitLong)
db, _, sqlDB := dbtestutil.NewDBWithSQLDB(t, dbtestutil.WithDumpOnFailure())
logger := slogtest.Make(t, &slogtest.Options{IgnoreErrors: true})
clk := quartz.NewMock(t)
now := clk.Now().UTC()
// Insert telemetry heartbeats.
err := db.InsertTelemetryLock(ctx, database.InsertTelemetryLockParams{
EventType: "aibridge_interceptions_summary",
PeriodEndingAt: now.Add(-25 * time.Hour), // should be purged
})
require.NoError(t, err)
err = db.InsertTelemetryLock(ctx, database.InsertTelemetryLockParams{
EventType: "aibridge_interceptions_summary",
PeriodEndingAt: now.Add(-23 * time.Hour), // should be kept
})
require.NoError(t, err)
err = db.InsertTelemetryLock(ctx, database.InsertTelemetryLockParams{
EventType: "aibridge_interceptions_summary",
PeriodEndingAt: now, // should be kept
})
require.NoError(t, err)
done := awaitDoTick(ctx, t, clk)
closer := dbpurge.New(ctx, logger, db, clk)
defer closer.Close()
<-done // doTick() has now run.
require.Eventuallyf(t, func() bool {
// We use an SQL queries directly here because we don't expose queries
// for deleting heartbeats in the application code.
var totalCount int
err := sqlDB.QueryRowContext(ctx, `
SELECT COUNT(*) FROM telemetry_locks;
`).Scan(&totalCount)
assert.NoError(t, err)
var oldCount int
err = sqlDB.QueryRowContext(ctx, `
SELECT COUNT(*) FROM telemetry_locks WHERE period_ending_at < $1;
`, now.Add(-24*time.Hour)).Scan(&oldCount)
assert.NoError(t, err)
// Expect 2 heartbeats remaining and none older than 24 hours.
t.Logf("eventually: total count: %d, old count: %d", totalCount, oldCount)
return totalCount == 2 && oldCount == 0
}, testutil.WaitShort, testutil.IntervalFast, "it should delete old telemetry heartbeats")
}