keith-turner opened a new pull request, #4563:
URL: https://github.com/apache/accumulo/pull/4563
The following changes were made as result of running Java Flight Recorder
repeatedly on the manager and tablet server while SplitMillionIT was running.
After these changes the following methods would not show up as much in the JFR
results.
* Sped up validation of FateId. When deleting 1 million tablets, 1 million
fate id are written and then read. Was seeing the regex for validation show up
when reading 1 million tablets w/ fate ids.
* Sped up getting children from ZooCache. The code related to service
locks was calling this.
* Sped up parsing of server locks by speeding up the UUID validation, that
is where it was spending most of its time.
* Sped up metadata constraint. Seeing conditional mutation metadata
updates spend a lot of time checking metadata constraints.
* Sped up the conditional check for absent walogs by removing the creation
of an empty TabletMetadata object
* Sped up SetEncodingIterator.encode by having a special case for size 1
and avoiding streams for size >1.
* Increased memory of manager and tsevers in SplitMillionIT because GC
pauses were being seen
* Pre split the metadata table in SplitMillionIT. This allowed the tablets
to spread across the two tablet servers. Pre splitting the metadata table
uncovered a bug. The add splits table operation would fail if metadata tables
it needed were not hosted. Fixed this bug.
* Made some other misc changes for little things that were seen in
profiling.
SplitMillionIT is now running faster, however it still does not seem as
fast as it used to be. Further investigation is needed. These changes are
mostly good general performance fixes. Can follow up wit more fixes as
investigation continues.
--
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.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
