What causes "reading props: %w" in weaviate/weaviate?

Read the wrapped underlying error to see whether it is a file-read, parse, or schema lookup failure

reading props: %w

Error message

reading props: %w

What it means

RunPrepareOnShard wraps failures from t.readPropsToReindex(rt) with the 'reading props' prefix. readPropsToReindex reads the reindex tracker / schema to determine which inverted-index properties must be re-indexed on this shard during the disk-heavy prep phase. A failure here means the property list could not be computed, so prep cannot proceed.

Source

Thrown at adapters/repos/db/inverted_reindex_task_generic.go:472

// the end of [runtimeSwap] after the atomic pointer flip.
func (t *ShardReindexTaskGeneric) RunPrepareOnShard(ctx context.Context, shard ShardLike) error {
	entry, err := t.enterDTMPhase(ctx, shard, "RunPrepareOnShard")
	if err != nil {
		return err
	}
	concreteShard, logger, rt := entry.shard, entry.logger, entry.rt

	// Idempotent fast-path: already merged means prep was already
	// completed (either by an earlier call in this process or by a
	// previous boot's runtimePrepare → markMerged).
	if rt.IsMerged() {
		logger.Debug("RunPrepareOnShard: already merged on disk; no-op")
		return nil
	}

	props, err := t.readPropsToReindex(rt)
	if err != nil {
		return fmt.Errorf("reading props: %w", err)
	}
	if len(props) == 0 {
		return fmt.Errorf("no props found for prep on shard %q", concreteShard.Name())
	}

	if err := t.ensureReindexBucketsLoadedForSwap(ctx, logger, concreteShard, props); err != nil {
		return fmt.Errorf("ensure buckets loaded: %w", err)
	}

	return t.runtimePrepare(ctx, logger, shard, rt, props)
}

// dtmPhaseEntry is the shared entry state of the DTM-scheduler-driven phase
// callbacks (RunPrepareOnShard / RunSwapOnShard).
type dtmPhaseEntry struct {
	shard  *Shard
	logger logrus.FieldLogger
	rt     reindexTracker

View on GitHub (pinned to 75aa4b6d11)

Solutions

Read the wrapped underlying error to see whether it is a file-read, parse, or schema lookup failure
Check the shard directory's migration/tracker files are readable and not corrupt; restore from a healthy node or backup if corrupt
Verify the class still exists in the schema and the migration is still applicable; cancel stale migrations if the class changed
Re-run RunPrepareOnShard — the merge phase is idempotent via sentinels

Defensive patterns

Strategy: try-catch

Validate before calling

// Pre-check: tracker file readable and class exists in schema
if _, err := os.Stat(shardDir + "/reindexed.mig"); err != nil { /* missing tracker: re-run earlier phases */ }
if !schema.ClassExists(className) { return fmt.Errorf("class %s gone; cancel migration", className) }

Try / catch

if err := task.RunPrepareOnShard(ctx, shard); err != nil {
	if strings.Contains(err.Error(), "reading props") {
		logger.Errorf("tracker/schema read failure on %s: %v", shard.Name(), err)
		// inspect tracker files, restore or re-trigger migration
	}
	return err
}

Prevention

Avoid concurrent schema edits (property removal) during active migrations
Back up the shard directory before manual restores so .mig tracker files stay intact
Ensure file permissions are consistent when copying shard dirs between nodes
Verify phase ordering: never call prep before reindex completed

When it happens

Trigger: Calling RunPrepareOnShard (directly or via the DTM scheduler paths dispatchMatrixDriveToMerged / dispatchMatrixDriveToTidied) when readPropsToReindex returns an error — typically because the reindex tracker's stored migration state is unreadable/corrupt or the schema lookup for the class fails.

Common situations: Corrupt .mig sentinel/tracker files on disk after a crash; schema class deleted concurrently while a migration is in flight; permission problems reading the tracker file after a restore/backup copy.

Related errors

AI-assisted analysis of weaviate/weaviate@75aa4b6d11 (2026-09-04). Data as JSON: /api/errors/f0fd4cebc4f46fe2. Report an issue: GitHub.