Fix panic when dealing with missing mapping ID #3188
Conversation
I have not gotten to the bottom of why the mapping ID is missing when maxNodes!=0. This is presenting a workaround until we full understand the issue. Fixes grafana#3164
|
@kolesnikovae I was able to reproduce the panic consistently with and this PR fixes it. Have a look at this code, the block is in a shared bucket that you should be able to access after being logged in: simonswine@3cf4015 All boils down to a single series |
|
How easy would it be to add a test that covers this specific case? |
There was a problem hiding this comment.
Great job!
I realize that the root cause of the problem lies a bit deeper – the profile that causes the issue is malformed: it has a single sample with the sentinel stack trace ID (0). Temporary fix: #3193. I'll investigate this further: #3195. Apart from the read path fixes, I think we should make sure such profiles do get into the block: #3194
This specific issue is caused by the fact that we do not expect that a profile has samples and does not have a mapping, therefore we reference the default one. Now that we rely on the mappings (containing the version and the git ref), I think we should always create a stub for truncated nodes – to be fixed separately: #3196.
I have not gotten to the bottom of why the mapping ID is missing when
maxNodes!=0.
This is presenting a workaround until we full understand the issue.
Fixes #3164