danny0405 commented on code in PR #12982:
URL: https://github.com/apache/hudi/pull/12982#discussion_r2024299129
##########
hudi-common/src/main/java/org/apache/hudi/common/table/view/RocksDbBasedFileSystemView.java:
##########
@@ -269,41 +269,51 @@ protected Option<Pair<String, CompactionOperation>>
getPendingLogCompactionOpera
@Override
protected boolean isPartitionAvailableInStore(String partitionPath) {
- String lookupKey = schemaHelper.getKeyForPartitionLookup(partitionPath);
- Serializable obj =
rocksDB.get(schemaHelper.getColFamilyForStoredPartitions(), lookupKey);
- return obj != null;
+ try {
+ readLock.lock();
+ String lookupKey = schemaHelper.getKeyForPartitionLookup(partitionPath);
+ Serializable obj =
rocksDB.get(schemaHelper.getColFamilyForStoredPartitions(), lookupKey);
+ return obj != null;
+ } finally {
+ readLock.unlock();
+ }
}
@Override
protected void storePartitionView(String partitionPath,
List<HoodieFileGroup> fileGroups) {
- LOG.info("Resetting and adding new partition ({}) to ROCKSDB based
file-system view at {}, Total file-groups={}",
- partitionPath, config.getRocksdbBasePath(), fileGroups.size());
-
- String lookupKey = schemaHelper.getKeyForPartitionLookup(partitionPath);
- rocksDB.delete(schemaHelper.getColFamilyForStoredPartitions(), lookupKey);
-
- // First delete partition views
- rocksDB.prefixDelete(schemaHelper.getColFamilyForView(),
- schemaHelper.getPrefixForSliceViewByPartition(partitionPath));
- rocksDB.prefixDelete(schemaHelper.getColFamilyForView(),
- schemaHelper.getPrefixForDataFileViewByPartition(partitionPath));
-
- // Now add them
- fileGroups.forEach(fg ->
- rocksDB.writeBatch(batch ->
- fg.getAllFileSlicesIncludingInflight().forEach(fs -> {
- rocksDB.putInBatch(batch, schemaHelper.getColFamilyForView(),
schemaHelper.getKeyForSliceView(fg, fs), fs);
- fs.getBaseFile().ifPresent(df ->
- rocksDB.putInBatch(batch,
schemaHelper.getColFamilyForView(), schemaHelper.getKeyForDataFileView(fg, fs),
df)
- );
- })
- )
- );
+ try {
+ writeLock.lock();
Review Comment:
> Previously, I was suggesting to make the map impls thread safe but it was
suggested to solve it at the view layer instead.
Did you have benchmark numbers for the synchronized disk map?
Not sure if it makes sense if we just put the write lock in the direct
modification method such as `#ensurePartitionsLoadedCorrectly`, and remove the
read lock from it's invoker. The read lock is only for read APIs that does not
do any modification to the fs state.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
