Improve seekable_format/examples/parallel_compression #4382
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
This passes make flags, such as `-jN` for building in parallel, to the underlying make.
Some of these examples are intended to be parallel, and don't make sense to link against single-threaded libzstd. The filename of mt and nomt libzstd are identical, so it's still possible to link against the single-threaded one, just harder.
Previously, parallel_compression would only handle each job's results after ALL jobs were successfully queued. This caused all src/dst buffers to remain in memory until then! It also polled to check whether a job completed, which is racy without any memory barrier. Now, we flush results as a side effect of completing a job. Completed frames are placed in an ordered linked-list, and any eligible frames are flushed. This may be zero or multiple frames, depending on the order in which jobs finish. This design also makes it simple to support streaming input, so that is now available. Just pass `-` as the filename, and stdin/stdout will be used for I/O.
There was no memory barrier between writing and reading `done`, which would allow reordering to cause races. With so little data to handle after each job completes, we might as well just join.
Use ulimit to fail the test if we use O(filesize) memory, rather than O(threads).
Building lz4 as root was causing `make clean` to fail with permission errors. We used to have to install lz4 from source back in Ubuntu 14.04, but nowadays the installed lz4 is fine. Get rid of ancient helpers and cruft!
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Outdated suggestions cannot be applied.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.
Change the design of parallel_compression, to allow it to actually work in parallel without blowing up memory usage. Fixes #3980.
In the new approach, we don't wait for all jobs to be accepted by the queue before writing any results. Instead, as each job finishes, it adds its result to an ordered-linked-list, and then writes (and frees) as many results as are available. We use a mutex for synchronization, so this should no longer be racy. This approach also conveniently allows streaming input, since we no longer need to know the number of frames ahead of time.
An added test enforces that parallel_compression doesn't use excessive memory.
This doesn't change anything about the main zstd program, so we should leave #3980 open, or open a new issue about zstd CLI seekability.
A variety of other fixes were necessary to get test passing, or were obvious cleanups once this was complete. Each is a separate commit.