correctly handle hitting MaxTabletsScheduled in drain #5845
Conversation
|
⚪
🟡
*please be aware that the difference is based on comparing your commit and the last completed build from the post-commit, check comparation |
|
⚪
🟢
*please be aware that the difference is based on comparing your commit and the last completed build from the post-commit, check comparation |
|
⚪
🟢
*please be aware that the difference is based on comparing your commit and the last completed build from the post-commit, check comparation |
|
⚪
🔴
*please be aware that the difference is based on comparing your commit and the last completed build from the post-commit, check comparation |
ydb/core/mind/hive/hive_impl.cpp
Outdated
| TBestNodeResult result = FindBestNode(*tablet); | ||
| if (result.BestNode == nullptr) { | ||
| Execute(CreateRestartTablet(tabletId)); | ||
| } else if (result.BestNode != tablet->Node) { | ||
| if (IsTabletMoveExpedient(*tablet, *result.BestNode)) { | ||
| Execute(CreateRestartTablet(tabletId)); | ||
| } | ||
| } |
There was a problem hiding this comment.
The checks here are pointless - we kill the tablet if there is no node for it, but leave it if there is a different node with the move not being expedient. Also the checks did not guarantee that the tablet would be restarted on a different node - because things might change between this moment and the restart tablet transaction. Since I have to touch this code, removing them seems best - unconditional tablet restart is reasonable for Kick, that's what people use it for.
|
⚪
🟢
*please be aware that the difference is based on comparing your commit and the last completed build from the post-commit, check comparation |
|
⚪
🟢
*please be aware that the difference is based on comparing your commit and the last completed build from the post-commit, check comparation |
Changelog entry
fix a bug where tablet drain could finish prematurely
Changelog category
Additional information
...