[fix][tiered-storage] Don't cleanup data when offload met Metastore e…

…xception (#17512)

* [fix][tiered-storage] Don't cleanup data when offload met BadVersion
---

*Motivation*

There have two ways that will cause the offload data cleanup. One is met
offload conflict exception, and another is completeLedgerInfoForOffloaded
reaches max retry time and throws zookeeper exceptions.

We retry the zookeeper operation on connection loss exception. We should
be careful about this exception, because we may loss data if the metadata
update successfully.

When a MetaStore exception happens, we can not make sure the metadata update is
failed or not. Because we have a retry on the connection loss, it is
possible to get a BadVersion or other exception after retrying.

So we don't clean up the data if this happens.

*Modification*

- don't delete data if has meta store exception

* log error when skip deleting

* improve logs

(cherry picked from commit c2588ba)

managed-ledger/src/main/java/org/apache/bookkeeper/mledger/impl/ManagedLedgerImpl.java

@@ -2902,8 +2902,21 @@ private void offloadLoop(CompletableFuture<PositionImpl> promise, Queue<LedgerIn
                                            scheduledExecutor, name)
                             .whenComplete((ignore2, exception) -> {
                                     if (exception != null) {
-                                        log.error("[{}] Failed to offload data for the ledgerId {}",
+                                        Throwable e = FutureUtil.unwrapCompletionException(exception);
+                                        if (e instanceof MetaStoreException) {
+                                            // When a MetaStore exception happens, we can not make sure the metadata
+                                            // update is failed or not. Because we have a retry on the connection loss,
+                                            // it is possible to get a BadVersion or other exception after retrying.
+                                            // So we don't clean up the data if it has metadata operation exception.
+                                            log.error("[{}] Failed to update offloaded metadata for the ledgerId {}, "
+                                                    + "the offloaded data will not be cleaned up",
                                                 name, ledgerId, exception);
+                                            return;
+                                        } else {
+                                            log.error("[{}] Failed to offload data for the ledgerId {}, "
+                                                    + "clean up the offloaded data",
+                                                name, ledgerId, exception);
+                                        }
                                         cleanupOffloaded(
                                             ledgerId, uuid,
                                             driverName, driverMetadata,