thanos-io/thanos · warning

close reader

Error message

close reader

What it means

After successfully reading the header bytes, the code closes the object-storage range reader and wraps a non-nil rc.Close() error as "close reader". A Close failure here means the HTTP response body could not be cleanly drained/closed (connection teardown error); the header data itself was read fine.

Solutions

1. Retry the whole WriteBinary — the data was read successfully and this is a teardown-only failure.
2. Tune keep-alive/idle-connection settings in the objstore HTTP config to avoid reset sockets.
3. If a custom objstore.BucketReader wrapper is in use, ensure Close only surfaces genuine close errors and drains the body first.
4. If this error appears persistently with one provider, update the object-store client/Thanos version for known close-error handling fixes.

Example fix

// before: fail on any close error
if err := rc.Close(); err != nil {
    return nil, 0, errors.Wrap(err, "close reader")
}
// after (caller side): treat close failure as non-fatal when payload already read
if err := rc.Close(); err != nil {
    log.Warn("index header reader close failed (data already read)", "err", err)
}

// Go: header bytes are already in memory, so a close error can often be tolerated
if err := rebuild(); err != nil {
    if strings.Contains(err.Error(), "close reader") {
        log.Warn("non-fatal close error after reading index header; continuing", "err", err)
        return nil // fallback: data was fully read
    }
    return err
}

When it happens

Trigger: WriteBinary -> newChunkedIndexReader -> rc.Close() returning an error when closing the ranged GET response body: underlying connection reset during close, provider client returning close-time errors (e.g. some s3/gcs client wrappers surface EOF here), or double-close on a wrapped reader.

Common situations: Network blip exactly at connection teardown; custom objstore wrappers whose Close propagates body-read leftovers as errors; aggressive connection reuse with broken keep-alive sockets.

Related errors

• read meta file
• get header from object storage of
• close toc reader
• copy symbols
• copy posting offsets

AI-assisted analysis of thanos-io/thanos@35b8b99117 (2026-09-07). Data as JSON: /api/errors/a427419368a883eb. Report an issue: GitHub.

Appendix: source

Thrown at pkg/block/indexheader/binary_reader.go:200

	indexFilepath := filepath.Join(id.String(), block.IndexFilename)
	attrs, err := bkt.Attributes(ctx, indexFilepath)
	if err != nil {
		return nil, 0, errors.Wrapf(err, "get object attributes of %s", indexFilepath)
	}

	rc, err := bkt.GetRange(ctx, indexFilepath, 0, index.HeaderLen)
	if err != nil {
		return nil, 0, errors.Wrapf(err, "get TOC from object storage of %s", indexFilepath)
	}

	b, err := io.ReadAll(rc)
	if err != nil {
		runutil.CloseWithErrCapture(&err, rc, "close reader")
		return nil, 0, errors.Wrapf(err, "get header from object storage of %s", indexFilepath)
	}

	if err := rc.Close(); err != nil {
		return nil, 0, errors.Wrap(err, "close reader")
	}

	if m := binary.BigEndian.Uint32(b[0:4]); m != index.MagicIndex {
		return nil, 0, errors.Errorf("invalid magic number %x for %s", m, indexFilepath)
	}

	version := int(b[4:5][0])

	if version != index.FormatV1 && version != index.FormatV2 {
		return nil, 0, errors.Errorf("not supported index file version %d of %s", version, indexFilepath)
	}

	ir := &chunkedIndexReader{
		ctx:  ctx,
		path: indexFilepath,
		size: uint64(attrs.Size),
		bkt:  bkt,
	}

View on GitHub (pinned to 35b8b99117)