umesh9794 commented on code in PR #10549:
URL: https://github.com/apache/ozone/pull/10549#discussion_r3449968387
##########
hadoop-ozone/recon/src/main/java/org/apache/hadoop/ozone/recon/persistence/ContainerHealthSchemaManager.java:
##########
@@ -205,28 +223,76 @@ public void replaceUnhealthyContainerRecordsAtomically(
private int deleteScmStatesForContainers(DSLContext dslContext,
List<Long> containerIds) {
+
+ // Sort and de-duplicate so contiguous runs are detectable regardless of
the
+ // order in which the caller supplied the ids.
+ List<Long> sortedIds = containerIds.stream()
+ .distinct()
+ .sorted()
+ .collect(Collectors.toList());
+
int totalDeleted = 0;
+ List<Long> inClauseBatch = new ArrayList<>(MAX_IN_CLAUSE_CHUNK_SIZE);
+
+ int i = 0;
+ while (i < sortedIds.size()) {
+ int runStart = i;
+ while (i + 1 < sortedIds.size()
+ && sortedIds.get(i + 1) == sortedIds.get(i) + 1) {
+ i++;
+ }
- for (int from = 0; from < containerIds.size(); from +=
MAX_IN_CLAUSE_CHUNK_SIZE) {
- int to = Math.min(from + MAX_IN_CLAUSE_CHUNK_SIZE, containerIds.size());
- List<Long> chunk = containerIds.subList(from, to);
-
- int deleted = dslContext.deleteFrom(UNHEALTHY_CONTAINERS)
- .where(UNHEALTHY_CONTAINERS.CONTAINER_ID.in(chunk))
- .and(UNHEALTHY_CONTAINERS.CONTAINER_STATE.in(
- UnHealthyContainerStates.MISSING.toString(),
- UnHealthyContainerStates.EMPTY_MISSING.toString(),
- UnHealthyContainerStates.UNDER_REPLICATED.toString(),
- UnHealthyContainerStates.OVER_REPLICATED.toString(),
- UnHealthyContainerStates.MIS_REPLICATED.toString(),
- UnHealthyContainerStates.NEGATIVE_SIZE.toString(),
- UnHealthyContainerStates.REPLICA_MISMATCH.toString()))
- .execute();
- totalDeleted += deleted;
+ if (i - runStart + 1 >= MIN_RANGE_DELETE_RUN) {
+ // Long contiguous run: a single constant-size BETWEEN removes the
whole
+ // range without straining Derby's per-statement bytecode budget.
+ totalDeleted += deleteScmStatesByRange(dslContext,
+ sortedIds.get(runStart), sortedIds.get(i));
Review Comment:
I was trying to keep the `DELETE` statements with `BETWEEN` and `IN` clause
separate to issue 2 `DELETE` statements. So in the worst case it falls back to
the same IN chunking the code used before this change, so it's never worse than
the old baseline of ceil(n / 1000). In best case it will issue just 1 statement
for a fully contiguous block.
As the comments says above combining both might overflow Derby's
per-statement bytecode limit (`Bytecode per method`).
If you prefer, I can combine it both into one. Thanks!
--
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]
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
