Periodically purge old database entries

kylecarbs · kylecarbs · commit 5a0b15dd7d39 · 2023-03-23T01:38:28.000Z
diff --git a/cli/server.go b/cli/server.go
@@ -65,6 +65,7 @@ import (
 	"github.com/coder/coder/coderd/autobuild/executor"
 	"github.com/coder/coder/coderd/database"
 	"github.com/coder/coder/coderd/database/dbfake"
+	"github.com/coder/coder/coderd/database/dbpurge"
 	"github.com/coder/coder/coderd/database/migrations"
 	"github.com/coder/coder/coderd/devtunnel"
 	"github.com/coder/coder/coderd/gitauth"
@@ -993,6 +994,10 @@ flags, and YAML configuration. The precedence is as follows:
 			shutdownConnsCtx, shutdownConns := context.WithCancel(ctx)
 			defer shutdownConns()
 
+			// Ensures that old database entries are cleaned up over time!
+			purger := dbpurge.New(ctx, logger, options.Database)
+			defer purger.Close()
+
 			// Wrap the server in middleware that redirects to the access URL if
 			// the request is not to a local IP.
 			var handler http.Handler = coderAPI.RootHandler
diff --git a/coderd/database/dbauthz/system.go b/coderd/database/dbauthz/system.go
@@ -280,6 +280,13 @@ func (q *querier) DeleteOldWorkspaceAgentStats(ctx context.Context) error {
 	return q.db.DeleteOldWorkspaceAgentStats(ctx)
 }
 
+func (q *querier) DeleteOldWorkspaceAgentStartupLogs(ctx context.Context) error {
+	if err := q.authorizeContext(ctx, rbac.ActionDelete, rbac.ResourceSystem); err != nil {
+		return err
+	}
+	return q.db.DeleteOldWorkspaceAgentStartupLogs(ctx)
+}
+
 func (q *querier) GetDeploymentWorkspaceAgentStats(ctx context.Context, createdAfter time.Time) (database.GetDeploymentWorkspaceAgentStatsRow, error) {
 	return q.db.GetDeploymentWorkspaceAgentStats(ctx, createdAfter)
 }
diff --git a/coderd/database/dbfake/databasefake.go b/coderd/database/dbfake/databasefake.go
@@ -4387,6 +4387,11 @@ func (q *fakeQuerier) DeleteLicense(_ context.Context, id int32) (int32, error)
 	return 0, sql.ErrNoRows
 }
 
+func (*fakeQuerier) DeleteOldWorkspaceAgentStartupLogs(_ context.Context) error {
+	// noop
+	return nil
+}
+
 func (q *fakeQuerier) GetUserLinkByLinkedID(_ context.Context, id string) (database.UserLink, error) {
 	q.mutex.RLock()
 	defer q.mutex.RUnlock()
diff --git a/coderd/database/dbpurge/dbpurge.go b/coderd/database/dbpurge/dbpurge.go
@@ -0,0 +1,64 @@
+package dbpurge
+
+import (
+	"context"
+	"errors"
+	"io"
+	"time"
+
+	"golang.org/x/sync/errgroup"
+
+	"cdr.dev/slog"
+	"github.com/coder/coder/coderd/database"
+)
+
+// New creates a new periodically purging database instance.
+// It is the caller's responsibility to call Close on the returned instance.
+//
+// This is for cleaning up old, unused resources from the database that take up space.
+func New(ctx context.Context, logger slog.Logger, db database.Store) io.Closer {
+	closed := make(chan struct{})
+	ctx, cancelFunc := context.WithCancel(ctx)
+	go func() {
+		defer close(closed)
+		ticker := time.NewTicker(24 * time.Hour)
+		defer ticker.Stop()
+		for {
+			select {
+			case <-ctx.Done():
+				return
+			case <-ticker.C:
+			}
+
+			var eg errgroup.Group
+			eg.Go(func() error {
+				return db.DeleteOldWorkspaceAgentStartupLogs(ctx)
+			})
+			eg.Go(func() error {
+				return db.DeleteOldWorkspaceAgentStats(ctx)
+			})
+			err := eg.Wait()
+			if err != nil {
+				if errors.Is(err, context.Canceled) {
+					return
+				}
+				logger.Error(ctx, "failed to purge old database entries", slog.Error(err))
+			}
+		}
+	}()
+	return &instance{
+		cancel: cancelFunc,
+		closed: closed,
+	}
+}
+
+type instance struct {
+	cancel context.CancelFunc
+	closed chan struct{}
+}
+
+func (i *instance) Close() error {
+	i.cancel()
+	<-i.closed
+	return nil
+}
diff --git a/coderd/database/dbpurge/dbpurge_test.go b/coderd/database/dbpurge/dbpurge_test.go
@@ -0,0 +1,26 @@
+package dbpurge_test
+
+import (
+	"context"
+	"testing"
+
+	"go.uber.org/goleak"
+
+	"github.com/stretchr/testify/require"
+
+	"cdr.dev/slog/sloggers/slogtest"
+	"github.com/coder/coder/coderd/database/dbfake"
+	"github.com/coder/coder/coderd/database/dbpurge"
+)
+
+func TestMain(m *testing.M) {
+	goleak.VerifyTestMain(m)
+}
+
+// Ensures no goroutines leak.
+func TestPurge(t *testing.T) {
+	t.Parallel()
+	purger := dbpurge.New(context.Background(), slogtest.Make(t, nil), dbfake.New())
+	err := purger.Close()
+	require.NoError(t, err)
+}
diff --git a/coderd/database/querier.go b/coderd/database/querier.go
diff --git a/coderd/database/queries.sql.go b/coderd/database/queries.sql.go
diff --git a/coderd/database/queries/workspaceagents.sql b/coderd/database/queries/workspaceagents.sql
@@ -128,6 +128,7 @@ INSERT INTO
 
 -- If an agent hasn't connected in the last 7 days, we purge it's logs.
 -- Logs can take up a lot of space, so it's important we clean up frequently.
+-- name: DeleteOldWorkspaceAgentStartupLogs :exec
 DELETE FROM workspace_agent_startup_logs WHERE agent_id IN
 	(SELECT id FROM workspace_agents WHERE last_connected_at IS NOT NULL
 		AND last_connected_at < NOW() - INTERVAL '7 day');
diff --git a/coderd/metricscache/metricscache.go b/coderd/metricscache/metricscache.go
@@ -9,7 +9,6 @@ import (
 
 	"golang.org/x/exp/maps"
 	"golang.org/x/exp/slices"
-	"golang.org/x/xerrors"
 
 	"github.com/google/uuid"
 
@@ -168,10 +167,6 @@ func countUniqueUsers(rows []database.GetTemplateDAUsRow) int {
 func (c *Cache) refreshTemplateDAUs(ctx context.Context) error {
 	//nolint:gocritic // This is a system service.
 	ctx = dbauthz.AsSystemRestricted(ctx)
-	err := c.database.DeleteOldWorkspaceAgentStats(ctx)
-	if err != nil {
-		return xerrors.Errorf("delete old stats: %w", err)
-	}
 
 	templates, err := c.database.GetTemplates(ctx)
 	if err != nil {

Original file line number	Diff line number	Diff line change
`@@ -280,6 +280,13 @@ func (q *querier) DeleteOldWorkspaceAgentStats(ctx context.Context) error {`
`280`	`280`	`return q.db.DeleteOldWorkspaceAgentStats(ctx)`
`281`	`281`	`}`
`282`	`282`
	`283`	`+func (q *querier) DeleteOldWorkspaceAgentStartupLogs(ctx context.Context) error {`
	`284`	`+ if err := q.authorizeContext(ctx, rbac.ActionDelete, rbac.ResourceSystem); err != nil {`
	`285`	`+ return err`
	`286`	`+ }`
	`287`	`+ return q.db.DeleteOldWorkspaceAgentStartupLogs(ctx)`
	`288`	`+}`
	`289`	`+`
`283`	`290`	`func (q *querier) GetDeploymentWorkspaceAgentStats(ctx context.Context, createdAfter time.Time) (database.GetDeploymentWorkspaceAgentStatsRow, error) {`
`284`	`291`	`return q.db.GetDeploymentWorkspaceAgentStats(ctx, createdAfter)`
`285`	`292`	`}`