Merge pull request #389 from gouthamve/wal-corruptio

Make sure WAL Repair can handle wrapped errors
This commit is contained in:
Fabian Reinartz 2018-09-19 08:47:24 +02:00 committed by GitHub
commit dfcb7d0d50
No known key found for this signature in database
GPG key ID: 4AEE18F83AFDEB23
2 changed files with 6 additions and 3 deletions

View file

@ -255,15 +255,17 @@ Loop:
// Repair attempts to repair the WAL based on the error.
// It discards all data after the corruption.
func (w *WAL) Repair(err error) error {
func (w *WAL) Repair(origErr error) error {
// We could probably have a mode that only discards torn records right around
// the corruption to preserve as data much as possible.
// But that's not generally applicable if the records have any kind of causality.
// Maybe as an extra mode in the future if mid-WAL corruptions become
// a frequent concern.
err := errors.Cause(origErr) // So that we can pick up errors even if wrapped.
cerr, ok := err.(*CorruptionErr)
if !ok {
return errors.New("cannot handle error")
return errors.Wrap(origErr, "cannot handle error")
}
if cerr.Segment < 0 {
return errors.New("corruption error does not specify position")

View file

@ -23,6 +23,7 @@ import (
"os"
"testing"
"github.com/pkg/errors"
"github.com/prometheus/tsdb/testutil"
)
@ -286,8 +287,8 @@ func TestWAL_Repair(t *testing.T) {
for r.Next() {
}
testutil.NotOk(t, r.Err())
testutil.Ok(t, w.Repair(r.Err()))
testutil.Ok(t, w.Repair(errors.Wrap(r.Err(), "err"))) // See https://github.com/prometheus/prometheus/issues/4603
sr, err = NewSegmentsReader(dir)
testutil.Ok(t, err)