busbey commented on a change in pull request #623: HBASE-22749: Distributed MOB
compactions
URL: https://github.com/apache/hbase/pull/623#discussion_r325901637
##########
File path:
hbase-server/src/main/java/org/apache/hadoop/hbase/mob/DefaultMobStoreCompactor.java
##########
@@ -183,105 +270,166 @@ protected boolean performCompaction(FileDetails fd,
InternalScanner scanner, Cel
boolean hasMore;
Path path = MobUtils.getMobFamilyPath(conf, store.getTableName(),
store.getColumnFamilyName());
byte[] fileName = null;
- StoreFileWriter mobFileWriter = null, delFileWriter = null;
- long mobCells = 0, deleteMarkersCount = 0;
+ StoreFileWriter mobFileWriter = null;
+ long mobCells = 0;
long cellsCountCompactedToMob = 0, cellsCountCompactedFromMob = 0;
long cellsSizeCompactedToMob = 0, cellsSizeCompactedFromMob = 0;
boolean finished = false;
+
ScannerContext scannerContext =
ScannerContext.newBuilder().setBatchLimit(compactionKVMax).build();
throughputController.start(compactionName);
- KeyValueScanner kvs = (scanner instanceof KeyValueScanner)?
(KeyValueScanner)scanner : null;
- long shippedCallSizeLimit = (long) numofFilesToCompact *
this.store.getColumnFamilyDescriptor().getBlocksize();
+ KeyValueScanner kvs = (scanner instanceof KeyValueScanner) ?
(KeyValueScanner) scanner : null;
+ long shippedCallSizeLimit =
+ (long) numofFilesToCompact *
this.store.getColumnFamilyDescriptor().getBlocksize();
+
+ MobCell mobCell = null;
try {
try {
// If the mob file writer could not be created, directly write the
cell to the store file.
mobFileWriter = mobStore.createWriterInTmp(new Date(fd.latestPutTs),
fd.maxKeyCount,
compactionCompression, store.getRegionInfo().getStartKey(), true);
fileName = Bytes.toBytes(mobFileWriter.getPath().getName());
} catch (IOException e) {
- LOG.warn("Failed to create mob writer, "
- + "we will continue the compaction by writing MOB cells
directly in store files", e);
+ // Bailing out
Review comment:
What's the reasoning for changing behavior here? If we have a normal store
writer and write all the cells in there, they'll get dumped back into a MOB
file on the next compaction pass right? (Just trying to work out if the
reasoning is a perf thing or a correctness thing)
----------------------------------------------------------------
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.
For queries about this service, please contact Infrastructure at:
[email protected]
With regards,
Apache Git Services
