the-other-tim-brown commented on code in PR #13269:
URL: https://github.com/apache/hudi/pull/13269#discussion_r2094161295
##########
hudi-client/hudi-client-common/src/main/java/org/apache/hudi/client/BaseHoodieWriteClient.java:
##########
@@ -940,46 +954,53 @@ public String startCommit(String actionType,
HoodieTableMetaClient metaClient) {
*/
public void startCommitWithTime(String instantTime) {
HoodieTableMetaClient metaClient = createMetaClient(true);
- startCommitWithTime(instantTime, metaClient.getCommitActionType(),
metaClient);
+ startCommitWithTime(Option.of(instantTime),
metaClient.getCommitActionType(), metaClient);
}
/**
* Completes a new commit time for a write operation
(insert/update/delete/insert_overwrite/insert_overwrite_table) with specified
action.
*/
public void startCommitWithTime(String instantTime, String actionType) {
HoodieTableMetaClient metaClient = createMetaClient(true);
- startCommitWithTime(instantTime, actionType, metaClient);
+ startCommitWithTime(Option.of(instantTime), actionType, metaClient);
}
/**
* Starts a new commit time for a write operation (insert/update/delete)
with specified action.
*/
- private void startCommitWithTime(String instantTime, String actionType,
HoodieTableMetaClient metaClient) {
+ private String startCommitWithTime(Option<String> providedInstantTime,
String actionType, HoodieTableMetaClient metaClient) {
if (needsUpgrade(metaClient)) {
// unclear what instant to use, since upgrade does have a given instant.
executeUsingTxnManager(Option.empty(), () -> tryUpgrade(metaClient,
Option.empty()));
}
CleanerUtils.rollbackFailedWrites(config.getFailedWritesCleanPolicy(),
HoodieTimeline.COMMIT_ACTION, () ->
tableServiceClient.rollbackFailedWrites(metaClient));
- LOG.info("Generate a new instant time: {} action: {}", instantTime,
actionType);
- // check there are no inflight restore before starting a new commit.
- HoodieTimeline inflightRestoreTimeline =
metaClient.getActiveTimeline().getRestoreTimeline().filterInflightsAndRequested();
- ValidationUtils.checkArgument(inflightRestoreTimeline.countInstants() == 0,
- () -> "Found pending restore in active timeline. Please complete the
restore fully before proceeding. As of now, "
- + "table could be in an inconsistent state. Pending restores: "
- +
Arrays.toString(inflightRestoreTimeline.getInstantsAsStream().map(HoodieInstant::requestedTime).toArray()));
-
- if (config.getFailedWritesCleanPolicy().isLazy()) {
- this.heartbeatClient.start(instantTime);
- }
+ txnManager.beginTransaction(Option.empty(), Option.empty());
+ String instantTime;
+ try {
+ instantTime = providedInstantTime.orElseGet(() ->
createNewInstantTime(false));
+ LOG.info("Generate a new instant time: {} action: {}", instantTime,
actionType);
+ // check there are no inflight restore before starting a new commit.
+ HoodieTimeline inflightRestoreTimeline =
metaClient.reloadActiveTimeline().getRestoreTimeline().filterInflightsAndRequested();
Review Comment:
I think this is a mistake but I am not sure. My reason for thinking it may
be a mistake is that there can be another writer that starts a restore between
generating the metaclient and its first timeline reload so you may miss the
restore. The question is whether this miss will cause any correctness issues or
if this will just waste some work, this is the part I am unsure about. If it is
just wasted work then we don't need the reload since that just forces more
listing in the general case of normal operations.
--
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]
