HADOOP-18221. Drains stream async before closing #4294
Closed
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
This is the the initial merge of the HADOOP-18028 S3A performance input stream. This patch on its own is incomplete and must be accompanied by all other commits with HADOOP-18028 in their git commit message. Consult the JIRA for that list Contributed by Bhalchandra Pandit.
…3A prefetching stream (apache#4115) Contributed by PJ Fanning.
Contributed by Ahmar Suhail
…ache#4212) Contributed by Monthon Klongklaew
|
💔 -1 overall
This message was automatically generated. |
|
@ahmarsuhail In Can we create a JIRA to add it after rebase? |
|
Thanks danny, I've created https://issues.apache.org/jira/browse/HADOOP-18230 |
|
💔 -1 overall
This message was automatically generated. |
|
@steveloughran this is ready for review now, would you be able to take a look? |
steveloughran
requested changes
May 18, 2022
There was a problem hiding this comment.
look at look at the changes in hadoop trunk s3a input stream in #2584 as a basis for this work. (i plan to rebase this branch this week, so you will have merge problems...sorry)
async draining does deliver speedups, but only if the amount of data to be read is "large enough". for small amounts of data, synchronous draining is lower overhead and guarantees the active http connection can be reused.
when i do there merge there will be an async drain threshold for this.
|
|
||
| Io.closeIgnoringIoException(this.inputStream); | ||
| Io.closeIgnoringIoException(this.obj); | ||
| } catch (Exception e) { |
There was a problem hiding this comment.
if this happens then the readl: raised an exception. the stream MUST be aborted to stop it being returned to the http connection pool, as its connection is probably broken
| public void run() { | ||
| try { | ||
|
|
||
| while(this.inputStream.read() >= 0) { |
There was a problem hiding this comment.
look at the changes in hadoop trunk s3a input stream here...it reads into a buffer for draining, and is marginally faster
| /** | ||
| * Drain task that is submitted to the future pool. | ||
| */ | ||
| private static class DrainTask implements Runnable { |
There was a problem hiding this comment.
- declare final to keep style checker happy
- i'd prefer to not use Runnable, instead completable futures., look at drainOrAbortHttpStream() and its use. at which point you can just pass in a function
| @@ -98,6 +106,7 @@ public S3File( | |||
| this.streamStatistics = streamStatistics; | |||
| this.changeTracker = changeTracker; | |||
| this.s3Objects = new IdentityHashMap<InputStream, S3Object>(); | |||
| this.futurePool = context.getFuturePool(); | |||
There was a problem hiding this comment.
if this is only for the drain, given the context is already stored, you can just get the pool when needed
asfgit
force-pushed
the
feature-HADOOP-18028-s3a-prefetch
branch
from
f38bbe2 to
b75b72b
Compare
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.
Description of PR
If close the prefetching input stream before prefetched blocks have finished reading the S3 input stream, the sdk repeatedly complains "Not all bytes were read from the S3ObjectInputStream". This happened on
S3AInputStreamas well, see this issue for more details.Closing the stream before draining it will abort the connection, so to allow for connection reuse we drain it asynchronously.
How was this patch tested?
Tested in eu-west-1 by running
mvn -Dparallel-tests -DtestsThreadCount=16 clean verify