This problem surfaced in impala integration tests
IMPALA-11592. TestLocalCatalogRetries.test_fetch_metadata_retry fails in S3 build
after the change
HADOOP-17461. Add thread-level IOStatistics Context
The actual GC race condition came with
HADOOP-18091. S3A auditing leaks memory through ThreadLocal references
The fix for this is, if our hypothesis is correct, in WeakReferenceMap.create()
where a strong reference to the new value is kept in a local variable
*and referred to later* so that the JVM will not GC it.
Along with the fix, extra assertions ensure that if the problem is not fixed,
applications will fail faster/more meaningfully.
Contributed by Steve Loughran.
part of HADOOP-18103.
While merging the ranges in CheckSumFs, they are rounded up based on the
value of checksum bytes size which leads to some ranges crossing the EOF
thus they need to be fixed else it will cause EOFException during actual reads.
Contributed By: Mukund Thakur
Follow-up to HADOOP-12020 Support configuration of different S3 storage classes;
S3 storage class is now set when buffering to heap/bytebuffers, and when
creating directory markers
Contributed by Monthon Klongklaew
HADOOP-16202 "Enhance openFile()" added asynchronous draining of the
remaining bytes of an S3 HTTP input stream for those operations
(unbuffer, seek) where it could avoid blocking the active
thread.
This patch fixes the asynchronous stream draining to work and so
return the stream back to the http pool. Without this, whenever
unbuffer() or seek() was called on a stream and an asynchronous
drain triggered, the connection was not returned; eventually
the pool would be empty and subsequent S3 requests would
fail with the message "Timeout waiting for connection from pool"
The root cause was that even though the fields passed in to drain() were
converted to references through the methods, in the lambda expression
passed in to submit, they were direct references
operation = client.submit(
() -> drain(uri, streamStatistics,
false, reason, remaining,
object, wrappedStream)); /* here */
Those fields were only read during the async execution, at which
point they would have been set to null (or even a subsequent read).
A new SDKStreamDrainer class peforms the draining; this is a Callable
and can be submitted directly to the executor pool.
The class is used in both the classic and prefetching s3a input streams.
Also, calling unbuffer() switches the S3AInputStream from adaptive
to random IO mode; that is, it is considered a cue that future
IO will not be sequential, whole-file reads.
Contributed by Steve Loughran.
The JournalNodeSyncer will include the local instance in syncing when using a bind host (e.g. 0.0.0.0). There is a mechanism that is supposed to exclude the local instance, but it doesn't recognize the meta-address as a local address.
Running with bind addresses set to 0.0.0.0, the JournalNodeSyncer will log attempts to sync with itself as part of the normal syncing rotation. For an HA configuration running 3 JournalNodes, the "other" list used by the JournalNodeSyncer will include 3 proxies.
Exclude bound local addresses, including the use of a wildcard address in the bound host configurations, while still allowing multiple instances on the same host.
Allow sync attempts with unresolved addresses, so that sync attempts can drive resolution as servers become available.
Backport.
Signed-off-by: stack <stack@apache.org>
Declares its compatibility with Spark's dynamic
output partitioning by having the stream capability
"mapreduce.job.committer.dynamic.partitioning"
Requires a Spark release with SPARK-40034, which
does the probing before deciding whether to
accept/rejecting instantiation with
dynamic partition overwrite set
This feature can be declared as supported by
any other PathOutputCommitter implementations
whose algorithm and destination filesystem
are compatible.
None of the S3A committers are compatible.
The classic FileOutputCommitter is, but it
does not declare itself as such out of our fear
of changing that code. The Spark-side code
will automatically infer compatibility if
the created committer is of that class or
a subclass.
Contributed by Steve Loughran.
This addresses an issue where the plugin's default classpath
for executing tests fails to include
org.junit.platform.launcher.core.LauncherFactory.
Contributed by: Steve Vaughan Jr
Use the existing DomainNameResolver to leverage the pluggable resolution framework. This provides a means to perform a reverse lookup if needed.
Update default implementation of DNSDomainNameResolver to protect against returning the IP address as a string from a cached value.
Co-authored-by: Steve Vaughan Jr <s_vaughan@apple.com>
Back port to branch-3.3, to avoid reconnecting to the old address after detecting that the address has been updated.
* Use a stable hashCode to allow safe IP addr changes
* Add test that updated address is used
Once the address has been updated, it will be used in future calls. Test verifies that a second request succeeds and that it uses the existing updated address instead of having to re-resolve.
Co-authored-by: Steve Vaughan Jr <s_vaughan@apple.com>
JobID.toString() and TaskID.toString() to only be called
when the IDs are not null.
This doesn't surface in MapReduce, but Spark SQL can trigger
in job abort, where it may invoke abortJob() with an
incomplete TaskContext.
This patch MUST be applied to branches containing
HADOOP-17833. "Improve Magic Committer Performance."
Contributed by Steve Loughran.
The name of the option to enable/disable thread level statistics is
"fs.iostatistics.thread.level.enabled";
There is also an enabled() probe in IOStatisticsContext which can
be used to see if the thread level statistics is active.
Contributed by Viraj Jasani
* HADOOP-18321.Fix when to read an additional record from a BZip2 text file split
Co-authored-by: Ashutosh Gupta <ashugpt@amazon.com> and Reviewed by Akira Ajisaka.
(cherry picked from commit a432925f74)
