pg_tde: harden archive/restore helpers to clean up tmpfs on failure #582
+231
−12
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
…ir if the operation fails
|
|
Collaborator
|
Thanks for the PR. This is for sure a real bug and we have reported it in our issue tracking system as PG-1932. As a C developer I am not really a fan of what the AI has coded so we will likely write our own patch for it which I hope you do not mind. Again thanks a lot for the bug report and the PR! |
Author
|
@jeltz thanks for looking into this. No, I don't mind at all. Would be curious to learn the proper fix! |
Collaborator
|
Closing this PR since development is moving to https://github.com/percona/pg_tde. Also I do remember this PR, we have just been very busy with work on PostgreSQL 18 support which is what caused us to split the repos. :) |
jeltz
added a commit
to jeltz/pg_tde
that referenced
this pull request
Nov 13, 2025
As reported by Roman Leventov in percona/postgres#582 we had an issue where we easily could leak tmeporary files and directories when on failure, especially when the wrapped restore or archive command fails. The solution in this commit is to move the cleanup of temporary files and directories to and atexit() callback plus add a signal handler for SIGINT and SIGTERM. The signal handler makes sure to kill itself after running cleanup so the parent process gets the right error. We have a tiny race condition between creating the temporary directory and registering the signal handler but that does not seem worth fixing since it would just leak a directory, not the WAL file, the race conditions being tiny.
jeltz
added a commit
to jeltz/pg_tde
that referenced
this pull request
Nov 13, 2025
As reported by Roman Leventov in percona/postgres#582 we had an issue where we easily could leak tmeporary files and directories when on failure, especially when the wrapped restore or archive command fails. The solution in this commit is to move the cleanup of temporary files and directories to and atexit() callback plus add a signal handler for SIGINT and SIGTERM. The signal handler makes sure to kill itself after running cleanup so the parent process gets the right error. We have a tiny race condition between creating the temporary directory and registering the signal handler but that does not seem worth fixing since it would just leak a directory, not the WAL file, the race conditions being tiny. With the current hardcoded path of /dev/shm and the TAp test framework testing this seems a pain so no tests are included.
jeltz
added a commit
to jeltz/pg_tde
that referenced
this pull request
Nov 13, 2025
As reported by Roman Leventov in percona/postgres#582 we had an issue where we easily could leak temporary files and directories when on failure, especially when the wrapped restore or archive command fails. The solution in this commit is to move the cleanup of temporary files and directories to and atexit() callback plus add a signal handler for SIGINT and SIGTERM. The signal handler makes sure to kill itself after running cleanup so the parent process gets the right error. We have a tiny race condition between creating the temporary directory and registering the signal handler but that does not seem worth fixing since it would just leak a directory, not the WAL file, the race conditions being tiny. With the current hardcoded path of /dev/shm and the TAP test framework testing this seems a pain so no tests are included.
jeltz
added a commit
to jeltz/pg_tde
that referenced
this pull request
Nov 13, 2025
As reported by Roman Leventov in percona/postgres#582 we had an issue where we easily could leak temporary files and directories when on failure, especially when the wrapped restore or archive command fails. The solution in this commit is to move the cleanup of temporary files and directories to and atexit() callback plus add a signal handler for SIGINT and SIGTERM. The signal handler makes sure to kill itself after running cleanup so the parent process gets the right error. We have a tiny race condition between creating the temporary directory and registering the signal handler but that does not seem worth fixing since it would just leak a directory, not the WAL file, the race conditions being tiny. With the current hardcoded path of /dev/shm and the TAP test framework testing this seems a pain so no tests are included.
jeltz
added a commit
to jeltz/pg_tde
that referenced
this pull request
Nov 13, 2025
As reported by Roman Leventov in percona/postgres#582 we had an issue where we easily could leak temporary files and directories when on failure, especially when the wrapped restore or archive command fails. The solution in this commit is to move the cleanup of temporary files and directories to and atexit() callback plus add a signal handler for SIGINT and SIGTERM. The signal handler makes sure to kill itself after running cleanup so the parent process gets the right error. We have a tiny race condition between creating the temporary directory and registering the signal handler but that does not seem worth fixing since it would just leak a directory, not the WAL file, the race conditions being tiny. With the current hardcoded path of /dev/shm and the TAP test framework testing this seems a pain so no tests are included.
Collaborator
|
@leventov Sorry for the silence but we have been very busy with the PostgreSQL 18 release which necessitated splitting the repository into two (this one which now only contains our custom hooks and https://github.com/percona/pg_tde which contains the actual extension). But now there is a PR, percona/pg_tde#467, for fixing this. It will sadly probably not make it to the first pg_tde release with PG 18 support (scheduled to be called 2.1) since code freeze for that has already happened but it will be released in 2.2, whenever that happens. |
Author
|
@jeltz thank you very much! |
jeltz
added a commit
to jeltz/pg_tde
that referenced
this pull request
Nov 13, 2025
As reported by Roman Leventov in percona/postgres#582 we had an issue where we easily could leak temporary files and directories when on failure, especially when the wrapped restore or archive command fails. The solution in this commit is to move the cleanup of temporary files and directories to and atexit() callback plus add a signal handler for SIGINT and SIGTERM. The signal handler makes sure to kill itself after running cleanup so the parent process gets the right error. We have a tiny race condition between creating the temporary directory and registering the signal handler but that does not seem worth fixing since it would just leak a directory, not the WAL file, the race conditions being tiny. With the current hardcoded path of /dev/shm and the TAP test framework testing this seems a pain so no tests are included.
jeltz
added a commit
to percona/pg_tde
that referenced
this pull request
Nov 13, 2025
As reported by Roman Leventov in percona/postgres#582 we had an issue where we easily could leak temporary files and directories when on failure, especially when the wrapped restore or archive command fails. The solution in this commit is to move the cleanup of temporary files and directories to and atexit() callback plus add a signal handler for SIGINT and SIGTERM. The signal handler makes sure to kill itself after running cleanup so the parent process gets the right error. We have a tiny race condition between creating the temporary directory and registering the signal handler but that does not seem worth fixing since it would just leak a directory, not the WAL file, the race conditions being tiny. With the current hardcoded path of /dev/shm and the TAP test framework testing this seems a pain so no tests are included.
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
3 participants
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
The pg_tde helper binaries decrypt/reencrypt WAL via a temporary file under TMPFS_DIRECTORY (/dev/shm). On very low‑memory systems this tmpfs can fill, and previous behavior left temporary files/directories behind on failures or abrupt termination, compounding ENOSPC and retries.
Changes:
SIGINT/SIGTERM (atexit() + signal handlers).
pg_tde_restore_encrypt.c): size = encrypted source file size + 4MB slack.pg_tde_archive_decrypt.c): read xlog_seg_size from pg_control (derived from DEST-PATHdata dir); fallback to 16MB if unavailable + 4MB slack.
FWIW, the initial condition that has led this failures was a programming mistake on my part rather than pg_tde's defect, but even then, I think it's better to harden the achirve/restore binaries so that they don't exacerbate the problem by cluttering tmpfs.
I've not tested this patch in a realistic environment yet (I indent to do that, but maybe only the next week or so), yet I would like to publish the patch already for review and feedback.
I'm not an expert in C, Postgres, or pgBackRest; the patch is authored by AI coding agent; but I've reviewed it and it makes sense to me.