Handle ENOSPC more cleanly #3513
Merged
Conversation
This file contains 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
dbutenhof
added
Server
API
webbnh
previously approved these changes
Aug 4, 2023
| controller = cm.archive_root / "ABC" | ||
| controller.mkdir() | ||
| (controller / source_tarball.name).write_text("Send in the clones") | ||
|
|
||
| # Attempting to create a dataset from the md5 file should result in | ||
| # a duplicate dataset error | ||
| with pytest.raises(DuplicateTarball) as exc: | ||
| cm.create(source_tarball) | ||
| assert exc.value.tarball == Dataset.stem(source_tarball) |
There was a problem hiding this comment.
The coverage report indicates that we do not test the case of a pre-existing MD5 file (without a pre-existing tarball).
There was a problem hiding this comment.
Not a new case, nor is the condition very realistic. I had kinda hoped to get this done today and I'm trying to finish something else.
There was a problem hiding this comment.
All true, but, realistic or not, we have code that checks for it, and, therefore, it arguably should be exercised by the tests. Whether it should be addressed in this PR or not is a totally separate question. (I'm just reporting what I found...and I did approve the PR....)
webbnh
previously approved these changes
Aug 4, 2023
There was a problem hiding this comment.
Looks great.
I just have the one question lingering from our previous exchange.
webbnh
previously approved these changes
Aug 7, 2023
PBENCH-1239 In Pbench ops review, after staging the latest `main` and with the intent of testing the new `Tarball.extract` on a large dataset, we pushed a >7Gb `uperf` tarball. This failed with an internal error, leaving a partial tarball copy in the `ARCHIVE` controller directory, revealing several related problems: 1. The cache manager was using `shlib.copy2`, which copies the tarball from the staging area into the archive tree. Because `nginx` also caches the entire data stream, this *triples* the storage requirements to upload a tarball. 2. On copy failure, the cache manager did not delete the partial file. 3. While the initial data stream save code handled an `ENOSPC` specially, after mapping trouble in Werkzeug it was reported as a "server internal error", which is not ideal. 4. The MD5 file write was not similarly protected: and while this is a small file and `ENOSPC` is unlikely, we should be prepared to handle it gracefully. This PR changes the cache manager to use `shlib.move` (which was the original intent) to avoid a third copy of the tarball. On failure, we unlink the file. Both the initial tarball and MD5 write handle `ENOSPC` and return HTTP status 413 (request entity too large), which is not a perfect mapping but a standard error code that Werkzeug can handle.
webbnh
previously approved these changes
Aug 7, 2023
webbnh
requested changes
Aug 7, 2023
ndokos
approved these changes
Aug 7, 2023
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
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.
PBENCH-1239
In Pbench ops review, after staging the latest
mainand with the intent of testing the newTarball.extracton a large dataset, we pushed a >7Gbuperftarball. This failed with an internal error, leaving a partial tarball copy in theARCHIVEcontroller directory, revealing several related problems:shlib.copy2, which copies the tarball from the staging area into the archive tree. Becausenginxalso caches the entire data stream, this triples the storage requirements to upload a tarball.ENOSPCspecially, after mapping trouble in Werkzeug it was reported as a "server internal error", which is not ideal.ENOSPCis unlikely, we should be prepared to handle it gracefully.This PR changes the cache manager to use
shlib.move(which was the original intent) to avoid a third copy of the tarball. On failure, we unlink the file. Both the initial tarball and MD5 write handleENOSPCand return HTTP status 413 (request entity too large), which is not a perfect mapping but a standard error code that Werkzeug can handle.