[FLINK-25994] Implement FileStoreExpire #17
Conversation
| private final ManifestList manifestList; | ||
| private final FileStoreScan scan; | ||
|
|
||
| public FileStoreExpireImpl( |
There was a problem hiding this comment.
I am thinking that we can provide three options:
- snapshot-expire.retention: retention time for snapshot changes
- snapshot-expire.num-retained
- snapshot-expire.check-interval
There was a problem hiding this comment.
Expire is an operation called by users (for example by connector sinks). snapshot-expire.check-interval should be checked by users, not by the operation itself.
As for retention time I'll implement this.
There was a problem hiding this comment.
Can you modify the caller in a separate PR too? StoreGlobalCommitter?
| * Deleted sst files in current snapshot. They can be safely deleted from file system if | ||
| * this snapshot expires. | ||
| */ | ||
| List<ManifestEntry> deletedFiles(); |
There was a problem hiding this comment.
This deletedFiles is misunderstanding.
I prefer to add a new method planChangeFiles(), it returns all files. The caller can compute physical deleted files.
| return; | ||
| } | ||
|
|
||
| // binary search for the last snapshot to expire |
There was a problem hiding this comment.
It is not cost effective to use dichotomous lookups (assuming there are enough surviving snapshots).
Just start traversing from the beginning.
| Set<Path> sstInUse = new HashSet<>(); | ||
| FileStorePathFactory.SstPathFactoryCache sstPathFactoryCache = | ||
| new FileStorePathFactory.SstPathFactoryCache(pathFactory); | ||
| for (ManifestEntry entry : scan.withSnapshot(nextSnapshot.id()).plan().files()) { |
There was a problem hiding this comment.
I can see large number of repeat reads. We can add a TODO to add manifest cache in scan.
There was a problem hiding this comment.
We can optimize this by expiring all snapshots at once so no repeated reads.
parent 2a62022 author Alibaba-HZY <daowu.hzy@cainiao.com> 1681226598 +0800 committer Alibaba-HZY <daowu.hzy@cainiao.com> 1681226783 +0800 # 这是一个 13 个提交的组合。 parent 2a62022 author Alibaba-HZY <daowu.hzy@cainiao.com> 1681226598 +0800 committer Alibaba-HZY <daowu.hzy@cainiao.com> 1681226730 +0800 # 这是一个 8 个提交的组合。tree 46928987599714b071489a7b6d4957049e6ded7a parent 2a62022 author Alibaba-HZY <daowu.hzy@cainiao.com> 1681226598 +0800 committer Alibaba-HZY <daowu.hzy@cainiao.com> 1681226598 +0800 [core] Add streaming read from option (apache#778) # 这是提交说明 apache#8： [core] add test (apache#778) # 这是提交说明 apache#11： Update paimon-core/src/main/java/org/apache/paimon/CoreOptions.java Co-authored-by: Nicholas Jiang <programgeek@163.com> # 这是提交说明 apache#12： Update paimon-core/src/main/java/org/apache/paimon/CoreOptions.java Co-authored-by: Nicholas Jiang <programgeek@163.com> # 这是提交说明 apache#13： Update paimon-core/src/main/java/org/apache/paimon/CoreOptions.java Co-authored-by: Nicholas Jiang <programgeek@163.com> # 这是提交说明 apache#15： Update paimon-flink/paimon-flink-common/src/main/java/org/apache/paimon/flink/AbstractFlinkTableFactory.java Co-authored-by: Nicholas Jiang <programgeek@163.com> # 这是提交说明 apache#16： Update paimon-flink/paimon-flink-common/src/main/java/org/apache/paimon/flink/AbstractFlinkTableFactory.java Co-authored-by: Nicholas Jiang <programgeek@163.com> # 这是提交说明 apache#17： Update paimon-flink/paimon-flink-common/src/main/java/org/apache/paimon/flink/AbstractFlinkTableFactory.java Co-authored-by: Nicholas Jiang <programgeek@163.com> # 这是提交说明 apache#18： [core] add test (apache#778) # 这是提交说明 apache#19： merger sdda 这是一个 3 个提交的组合。 [core] commit1 (apache#778) [core] commit2 (apache#778) [core] commit3 (apache#778) [core] do commit 1(apache#778) [core] do commit 2(apache#778) # 这是提交说明 apache#20： sdda 这是一个 3 个提交的组合。 [core] commit1 (apache#778) [core] commit2 (apache#778) [core] commit3 (apache#778)
Currently FileStoreExpire does not have an implementation. We need an implementation to clean up old snapshots and related files.