-
Notifications
You must be signed in to change notification settings - Fork 85
chore: add BidiAppendableUnbufferedWritableByteChannel #3224
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
chore: add BidiAppendableUnbufferedWritableByteChannel #3224
Conversation
f3d0fff
to
a69bad6
Compare
0858e8a
to
41da934
Compare
a69bad6
to
4d96b53
Compare
90bddd1
to
bd3e81f
Compare
4d96b53
to
365bce5
Compare
This will have tests added in a followup commit
bd3e81f
to
6d24a64
Compare
} | ||
|
||
@Override | ||
public long writeAndClose(ByteBuffer[] srcs, int offset, int length) throws IOException { |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Why isn't this just
long written = write(srcs, offset, length);
close();
return written;
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
The internalWrite
and write
methods are both non-blocking and may not consume all of the bytes present in srcs. However, close does need to block. So we loop here to only call close once all of the bytes have been successfully enqueued. Then close can do it's normal blocking waiting for the necessary ack from gcs.
## Description feat: *breaking behavior* rewrite Storage.blobAppendableUpload to be non-blocking and have improved throughput (#3231) Rewrite internals of BlobAppendableUpload to provide non-blocking write calls, and it take advantage of grpc async message handling. When `AppendableUploadWriteableByteChannel#write(ByteBuffer)` is called, an attempt will be made to enqueue the bytes in the outbound queue to GCS. If there is only enough room to partially consume the bytes provided in the `ByteBuffer` the write call will return early specifying the number of bytes actually consumed. As acknowledgements come in from gcs, enqueued messages will be evicted freeing space in the outbound queue. Thereby allowing more bytes to be consumed and enqueued. Given appendable objects are still in private preview I can't quote any metrics here, however preliminary benchmarking of several million objects across a range of sizes show across the board throughput improvments. Because the channel's write call is now non-blocking, if you want to block your application until the full buffer is consumed some new helper methods have been added in StorageChannelUtils to provide blocking behavior. A new method `MinFlushSizeFlushPolicy#withMaxPendingBytes(long)` has been added to allow limiting the number of pending outbound bytes. The default values is 16MiB, but can be configured lower if necessary. ## Release Notes BEGIN_COMMIT_OVERRIDE BEGIN_NESTED_COMMIT feat: *breaking behavior* rewrite Storage.blobAppendableUpload to be non-blocking and have improved throughput (#3231) END_NESTED_COMMIT BEGIN_NESTED_COMMIT feat: add StorageChannelUtils to provide helper methods to perform blocking read/write to/from non-blocking channels (#3231) END_NESTED_COMMIT BEGIN_NESTED_COMMIT feat: add MinFlushSizeFlushPolicy#withMaxPendingBytes(long) (#3231) END_NESTED_COMMIT BEGIN_NESTED_COMMIT fix: update BlobAppendableUploadConfig and FlushPolicy.MinFlushSizeFlushPolicy to default to 4MiB minFlushSize and 16MiB maxPendingBytes (#3249) END_NESTED_COMMIT BEGIN_NESTED_COMMIT fix: make FlushPolicy${Min,Max}FlushSizeFlushPolicy constructors private (#3217) END_NESTED_COMMIT END_COMMIMT_OVERRIDE ## Sub PRs This PR is made of up the following PRs, in sequence 1. #3217 2. #3218 3. #3219 4. #3220 5. #3221 6. #3222 7. #3223 8. #3224 9. #3225 10. #3226 11. #3227 12. #3228 13. #3229 14. #3230 15. #3235 16. #3236 17. #3241 18. #3242 19. #3246 20. #3248 21. #3249 22. #3252
Merged in #3231 |
This will have tests added in a followup commit