New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
c-deps: Investigate RocksDB DeleteRange bug #18371
Comments
Ah, the TiKV-folks talk about using |
I ported the test case in https://github.com/facebook/rocksdb/pull/2799/files to our code base yet I'm not seeing a failure. Sort of curious that we haven't seen any consistency check failures recently. We've been using |
Oh, looks like their test is missing a One mitigating factor is that it looks like an explicit snapshot created via I'm going to attempt to reproduce by convincing replica GC and rebalancing to create the correct intermingling of RocksDB operations. It is possible that one of the existing remove-and-re-add tests sees this bug, but a) we're not checking for it and b) we don't use |
Btw, the fix is relatively small and I've verified it fixes my reproduction. We might want to just apply it. |
Agreed, given the size of the patch I think we should just apply it. What are the other options for 1.1? Disabling |
Yeah, let's just apply it. |
Ok. I'll add it to #18374. |
Manually apply the fix to our RocksDB repo. The presence of the test will prevent upgrading to a version of RocksDB that does not contain the fix. Fixes cockroachdb#18371
Since the bug's been around a while, should we backport it to 1.0 too? |
Are we going to do a 1.0.7? Or should we squeeze it into 1.0.6? I can whip up a cherrypick quickly. |
I think if it's ready today we could put it in 1.0.6 (we haven't started qualifying a specific build yet) |
The RocksDB patch didn't apply cleanly to the version of RocksDB we're using on the |
Manually apply the fix to our RocksDB repo. The presence of the test will prevent upgrading to a version of RocksDB that does not contain the fix. Fixes cockroachdb#18371
The TiDB team found a data corruption bug in RocksDB (blog post, issue, PR). We need to determine whether we are affected by this; if so we need to patch in the fix for 1.1.
The bug relates to the
DeleteRange
operation and snapshots; we use both features and based on the test case in facebook/rocksdb#2799 I don't see any reason to assume that we're immune. The symptom is data that was deleted by DeleteRange could reappear.The text was updated successfully, but these errors were encountered: