pgaref commented on a change in pull request #1823:
URL: https://github.com/apache/hive/pull/1823#discussion_r567807669
##########
File path: common/src/java/org/apache/hadoop/hive/conf/HiveConf.java
##########
@@ -4509,7 +4509,7 @@ private static void populateLlapDaemonVarsSet(Set<String>
llapDaemonVarsSetLocal
"Minimum allocation possible from LLAP buddy allocator. Allocations
below that are\n" +
"padded to minimum allocation. For ORC, should generally be the same
as the expected\n" +
"compression buffer size, or next lowest power of 2. Must be a power
of 2."),
- LLAP_ALLOCATOR_MAX_ALLOC("hive.llap.io.allocator.alloc.max", "16Mb", new
SizeValidator(),
+ LLAP_ALLOCATOR_MAX_ALLOC("hive.llap.io.allocator.alloc.max", "4Mb", new
SizeValidator(),
Review comment:
The issue here is that LLAP_ALLOCATOR_MAX_ALLOC is also used as the ORC
Writer buffer size (thus the change).
Initial buffer size check was introduced in
[ORC-238](https://github.com/apache/orc/pull/171/files) even though it was only
applied when buffer size was enforced from table properties. Later, on ORC-1.6
this was enforced for the [Writer buffer size in
general](https://github.com/apache/orc/blob/0128f817b0ab28fa2d0660737234ac966f0f5c50/java/core/src/java/org/apache/orc/impl/WriterImpl.java#L171).
The max bufferSize argument can be up to 2^(3*8 - 1) -- meaning less than
8Mb and since we enforce the size to be power of 2 the next available is 4Mb.
The main question here is if there is a reason for the upper limit to be < 8
Mb (cc @prasanthj that might know more here) -- or if we should decouple the
two configuration (LLAP alloc and ORC Writer buffer size).
I believe the best thing to do for now is open a new Ticket to track this
(as this will either require more work on LLAP, or a new release on ORC) -- and
I do not expect this to cause any major issues until then. @mustafaiman what do
you think?
----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
For queries about this service, please contact Infrastructure at:
[email protected]
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
