Avoid change feed rewinds on shard moves #3711
Merged
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
nickva
force-pushed
the
check-for-moved-shards
branch
6 times, most recently
from
6e42f23 to
21ca354
Compare
nickva
force-pushed
the
check-for-moved-shards
branch
5 times, most recently
from
c617864 to
b8e648d
Compare
jaydoane
approved these changes
Aug 26, 2021
Contributor
jaydoane
left a comment
jaydoane
left a comment
There was a problem hiding this comment.
LGTM, with just some minor formatting nits, typos, and a few questions about possible name improvements.
| % Since we are doing a best-effort approach to match moved shards, | ||
| % tollerate and log errors. This should also handle cases when the | ||
| % cluster is partially upgraded, as some nodes will not have the | ||
| % newer get_uuid fabric_rpc handler. |
Contributor
There was a problem hiding this comment.
this is a very nice feature!
nickva
force-pushed
the
check-for-moved-shards
branch
4 times, most recently
from
aa179e2 to
7945755
Compare
When shards are moved to new nodes, and the user supplies a change sequence from the old shard map configuration, attempt to match missing nodes and ranges by inspecting current shard uuids in order to avoid rewinds. Previously, if a node and range was missing, we randomly picked a node in the appropriate range, so 1/3 of the time we might have hit the exact node, but 2/3 of the time we would end up with a complete changes feed rewind to 0. Unfortunately, this involves a fabric worker scatter gather operation to all shard copies. This should only happen when we get an old sequence. We rely on that happening rarely, mostly right after the shards moved, then users would get new sequence from the recent shard map.
nickva
force-pushed
the
check-for-moved-shards
branch
from
7945755 to
6e33c46
Compare
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
When shards are moved to new nodes, and the user supplies a change sequence from the old shard map configuration, attempt to match missing nodes and ranges by inspecting current shard uuids in order to avoid rewinds.
Previously, if a node and range was missing, we randomly picked a node in the appropriate range, so 1/3 of the time we might have hit the exact node, but 2/3 of the time we would end up with a complete changes feed rewind to 0.
Unfortunately, this involves a fabric worker scatter gather operation to all shard copies. This should only happen when we get an old sequence. We rely on that happening rarely, mostly right after the shards moved, then users would get new sequence from the recent shard map.