[fix][offload] Close all resources in BlobStoreBackedReadHandleImplV2.closeAsync

[iamsanjaymalakar]

Motivation

BlobStoreBackedReadHandleImplV2#closeAsync() closes indices and dataStreams sequentially inside a single try/catch. If closing one resource throws, the method exits early and the remaining resources are never closed. This can leak file descriptors and leave resources open when offload read handles are closed.

Modifications

• Update closeAsync() to attempt closing all OffloadIndexBlockV2 instances and all DataInputStreams.

• Collect the first IOException and attach subsequent failures via suppressed exceptions.

• Complete the returned CompletableFuture exceptionally if any close operation fails; otherwise mark the handle Closed and complete normally.

• doc

• doc-required

• doc-not-needed

• doc-complete

[github-actions]

@iamsanjaymalakar Please add the following content to your PR description and select a checkbox:

- [ ] `doc` <!-- Your PR contains doc changes -->
- [ ] `doc-required` <!-- Your PR changes impact docs and you will update later -->
- [ ] `doc-not-needed` <!-- Your PR changes do not impact docs -->
- [ ] `doc-complete` <!-- Docs have been already added -->

[codelipenghui]

And could you please also add tests to cover the changes from this PR, which can confirm the issue can be reproduced and avoid regression in the future.

[iamsanjaymalakar]

Thanks for the reviews! I addressed the feedback by making closeAsync() close all resources without early exit and ensuring the returned future is always completed (even on runtime exceptions). I didn’t add a test because resource-leak/file-lock tests tend to be nondeterministic and flaky across OS/CI environments.

[lhotari]

Please run "Personal CI" to validate such changes on your side. You can run Pulsar CI in your personal fork by following these instructions: https://pulsar.apache.org/contribute/personal-ci/. It's free.
The benefit is that you have full control of CI and you won't need to wait for maintainers to approve the CI workflow run to get CI feedback.

[lhotari]

One detail about BlobStoreBackedReadHandleImplV2 is that it's not used in Pulsar's production code.
The
org.apache.bookkeeper.mledger.LedgerOffloader#readOffloaded(long, org.apache.bookkeeper.mledger.proto.MLDataFormats.OffloadContext, java.util.Map<java.lang.String,java.lang.String>) method is not called from production code which would use BlobStoreBackedReadHandleImplV2.
This something that I noticed while fixing a flaky tests a while ago in #24331.
"PIP 76: Streaming Offload" hasn't been fully implemented.