[GitHub] [hudi] TengHuo commented on a diff in pull request #5677: [HUDI-4152] Flink offline compaction support compacting multi compaction plan at once

[GitBox <gi...@apache.org>]

TengHuo commented on code in PR #5677:
URL: https://github.com/apache/hudi/pull/5677#discussion_r1057479092


##########
hudi-flink-datasource/hudi-flink/src/main/java/org/apache/hudi/sink/compact/HoodieFlinkCompactor.java:
##########
@@ -218,74 +228,112 @@ private void compact() throws Exception {
       }
 
       // fetch the instant based on the configured execution sequence
-      HoodieTimeline timeline = table.getActiveTimeline().filterPendingCompactionTimeline();
-      Option<HoodieInstant> requested = CompactionUtil.isLIFO(cfg.compactionSeq) ? timeline.lastInstant() : timeline.firstInstant();
-      if (!requested.isPresent()) {
+      HoodieTimeline timeline = table.getActiveTimeline();
+      List<HoodieInstant> requested = ((CompactionPlanSelectStrategy) ReflectionUtils.loadClass(cfg.compactionPlanSelectStrategy))
+          .select(timeline.filterPendingCompactionTimeline(), cfg);
+      if (requested.isEmpty()) {
         // do nothing.
         LOG.info("No compaction plan scheduled, turns on the compaction plan schedule with --schedule option");
         return;
       }
 
-      String compactionInstantTime = requested.get().getTimestamp();
-
-      HoodieInstant inflightInstant = HoodieTimeline.getCompactionInflightInstant(compactionInstantTime);
-      if (timeline.containsInstant(inflightInstant)) {
-        LOG.info("Rollback inflight compaction instant: [" + compactionInstantTime + "]");
-        table.rollbackInflightCompaction(inflightInstant);
-        table.getMetaClient().reloadActiveTimeline();
-      }
+      List<String> compactionInstantTimes = requested.stream().map(HoodieInstant::getTimestamp).collect(Collectors.toList());
+      compactionInstantTimes.forEach(timestamp -> {
+        HoodieInstant inflightInstant = HoodieTimeline.getCompactionInflightInstant(timestamp);
+        if (timeline.containsInstant(inflightInstant)) {
+          LOG.info("Rollback inflight compaction instant: [" + timestamp + "]");
+          table.rollbackInflightCompaction(inflightInstant);
+          table.getMetaClient().reloadActiveTimeline();
+        }
+      });
 
-      // generate compaction plan
+      // generate timestamp and compaction plan pair
       // should support configurable commit metadata
-      HoodieCompactionPlan compactionPlan = CompactionUtils.getCompactionPlan(
-          table.getMetaClient(), compactionInstantTime);
-
-      if (compactionPlan == null || (compactionPlan.getOperations() == null)
-          || (compactionPlan.getOperations().isEmpty())) {
+      List<Pair<String, HoodieCompactionPlan>> compactionPlans = compactionInstantTimes.stream()
+          .map(timestamp -> {
+            try {
+              return Pair.of(timestamp, CompactionUtils.getCompactionPlan(table.getMetaClient(), timestamp));
+            } catch (IOException e) {
+              throw new HoodieException(e);
+            }
+          })
+          // reject empty compaction plan
+          .filter(pair -> !(pair.getRight() == null
+              || pair.getRight().getOperations() == null
+              || pair.getRight().getOperations().isEmpty()))
+          .collect(Collectors.toList());
+
+      if (compactionPlans.isEmpty()) {
         // No compaction plan, do nothing and return.
-        LOG.info("No compaction plan for instant " + compactionInstantTime);
+        LOG.info("No compaction plan for instant " + String.join(",", compactionInstantTimes));
         return;
       }
 
-      HoodieInstant instant = HoodieTimeline.getCompactionRequestedInstant(compactionInstantTime);
+      List<HoodieInstant> instants = compactionInstantTimes.stream().map(HoodieTimeline::getCompactionRequestedInstant).collect(Collectors.toList());
       HoodieTimeline pendingCompactionTimeline = table.getActiveTimeline().filterPendingCompactionTimeline();
-      if (!pendingCompactionTimeline.containsInstant(instant)) {
-        // this means that the compaction plan was written to auxiliary path(.tmp)
-        // but not the meta path(.hoodie), this usually happens when the job crush
-        // exceptionally.
-
-        // clean the compaction plan in auxiliary path and cancels the compaction.
-
-        LOG.warn("The compaction plan was fetched through the auxiliary path(.tmp) but not the meta path(.hoodie).\n"
-            + "Clean the compaction plan in auxiliary path and cancels the compaction");
-        CompactionUtil.cleanInstant(table.getMetaClient(), instant);
-        return;
+      for (HoodieInstant instant : instants) {
+        if (!pendingCompactionTimeline.containsInstant(instant)) {
+          // this means that the compaction plan was written to auxiliary path(.tmp)
+          // but not the meta path(.hoodie), this usually happens when the job crush
+          // exceptionally.
+          // clean the compaction plan in auxiliary path and cancels the compaction.
+          LOG.warn("The compaction plan was fetched through the auxiliary path(.tmp) but not the meta path(.hoodie).\n"
+              + "Clean the compaction plan in auxiliary path and cancels the compaction");
+          CompactionUtil.cleanInstant(table.getMetaClient(), instant);
+          return;

Review Comment:
   Hi @lanyuanxiaoyao 
   I'm reviewing `HoodieFlinkCompactor` code recently and I got confused here as well.
   
   As the comment above `// this means that the compaction plan was written to auxiliary path(.tmp)`, it said there could be a situation that a compaction plan found in `.aux` folder (not `.tmp`, it doesn't exist, should be a typo), but not found in Hudi timeline. However, according to the code, `instants` is from a list `compactionInstantTimes`, which is from another list `requested`. So the source of `instants` is also Hudi timeline, which is not matched with the comment.
   
   Then, I checked the original PR for implementing this `HoodieFlinkCompactor`, https://github.com/apache/hudi/pull/3046. It also loaded this `instant` from Hudi timeline, not `.aux` folder. So I guess this part of code should be deprecated now. Am I right?



-- 
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: commits-unsubscribe@hudi.apache.org

For queries about this service, please contact Infrastructure at:
users@infra.apache.org