[
https://issues.apache.org/jira/browse/CASSANDRA-14523?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16530085#comment-16530085
]
Benjamin Lerer commented on CASSANDRA-14523:
--------------------------------------------
Thanks for the patch.
I think we should take advantages of the virtual table patches to clean a bit
the Metric classes. I pushed a commit
[here|https://github.com/apache/cassandra/commit/7e8ec2f22c8d58351e72f550f811108fdc71a0a7]
to remove the duplicated code in {{ThreadPoolMetrics}} and {{SEPMetrics}}.
One common problem to expose the metrics through virtual tables when the
{{DataSet}} is build on demand is to be able to retrieve the set of metrics.
Due to that the current patch has to fetch the metric values through {{JMX}}
and change the visibility of some thread pools. To avoid that problem, we could
keep a concurrent {{DataSet}} in memory and force the metric classes to add the
column values to the tables. I have pushed a patch to add a concurrent
{{DataSet}}
[here|https://github.com/apache/cassandra/commit/98ff447ec580bf42cc7141a4a484f79ef7f664a9]
and one to use it in {{ThreadPoolMetrics}}
[here|https://github.com/apache/cassandra/commit/960563fdc865c9c06b66c4f29b161ffbb3d7213d].
I had too change a bit the {{DataSet}} API to be able to add the concurrent
{{DataSet}} but the new API for building the {{DataSet}} can be used for both
type of {{DataSets}}.
Adding the metrics for the {{ScheduledThreadPool}} is a good idea. I just
wonder if we should not simply add some real metrics for them. We could do that
by replacing the {{ScheduledThreadPoolExecutor}} from {{SSTableReader}} by a
{{DebuggableScheduledThreadPoolExecutor}} and by having the
{{DebuggableScheduledThreadPoolExecutor}} class creating a metric for itself.
The advantage of that approach is that it will also be possible to monitor
those thread pools through {{nodetool}} and {{JMX}}. Ideally, I think it should
be done in another ticket.
I noticed that the patch reformat the name of the pools for exposing them in
the virtual table. We should probably keep the original name as it is a bit
confusing if other tools and logs display a different name.
It is not directly related to this patch but more a general remark about
Virtual Tables: I wonder if we should not use a {{LocalPartitioner}} for them.
Keeping the partition ordered by their partition key natural order is in my
opinion an advantage for the virtual tables. ([~iamaleksey] do you see any
problem with that?)
I pushed some commits
[here|https://github.com/apache/cassandra/compare/trunk...blerer:14523-trunk-review]
to cover all the things I mentioned.
What do you think [~cnlwsu]?
> Thread pool stats virtual table
> -------------------------------
>
> Key: CASSANDRA-14523
> URL: https://issues.apache.org/jira/browse/CASSANDRA-14523
> Project: Cassandra
> Issue Type: New Feature
> Reporter: Chris Lohfink
> Assignee: Chris Lohfink
> Priority: Minor
> Labels: virtual-tables
>
> Expose the thread pools like in status logger/tpstats. Additionally be nice
> to include the scheduled executor pools that are currently unmonitored.
> {code:java}
> cqlsh> select * from system_views.thread_pools;
> thread_pool | active | active_max | completed | pending
> | tasks_blocked | total_blocked
> ----------------------------------+--------+------------+-----------+---------+---------------+---------------
> anti_entropy_stage | 0 | 1 | 0 | 0
> | 0 | 0
> cache_cleanup_executor | 0 | 1 | 0 | 0
> | 0 | 0
> compaction_executor | 0 | 4 | 41 | 0
> | 0 | 0
> counter_mutation_stage | 0 | 32 | 0 | 0
> | 0 | 0
> gossip_stage | 0 | 1 | 0 | 0
> | 0 | 0
> hints_dispatcher | 0 | 2 | 0 | 0
> | 0 | 0
> internal_response_stage | 0 | 8 | 0 | 0
> | 0 | 0
> memtable_flush_writer | 0 | 2 | 5 | 0
> | 0 | 0
> memtable_post_flush | 0 | 1 | 20 | 0
> | 0 | 0
> memtable_reclaim_memory | 0 | 1 | 5 | 0
> | 0 | 0
> migration_stage | 0 | 1 | 0 | 0
> | 0 | 0
> misc_stage | 0 | 1 | 0 | 0
> | 0 | 0
> mutation_stage | 0 | 32 | 247 | 0
> | 0 | 0
> native_transport_requests | 1 | 128 | 28 | 0
> | 0 | 0
> pending_range_calculator | 0 | 1 | 2 | 0
> | 0 | 0
> per_disk_memtable_flush_writer_0 | 0 | 2 | 5 | 0
> | 0 | 0
> read_repair_stage | 0 | 8 | 0 | 0
> | 0 | 0
> read_stage | 0 | 32 | 13 | 0
> | 0 | 0
> repair_task | 0 | 2147483647 | 0 | 0
> | 0 | 0
> request_response_stage | 0 | 8 | 0 | 0
> | 0 | 0
> sampler | 0 | 1 | 0 | 0
> | 0 | 0
> scheduled_fast_tasks | 0 | 2147483647 | 1398 | 1
> | 0 | 0
> scheduled_heartbeat | 0 | 2147483647 | 14 | 1
> | 0 | 0
> scheduled_hotness_tracker | 0 | 2147483647 | 0 | 1
> | 0 | 0
> scheduled_non_periodic_tasks | 0 | 2147483647 | 10 | 0
> | 0 | 0
> scheduled_optional_tasks | 0 | 2147483647 | 5 | 8
> | 0 | 0
> scheduled_summary_builder | 0 | 2147483647 | 0 | 1
> | 0 | 0
> scheduled_tasks | 0 | 2147483647 | 194 | 74
> | 0 | 0
> secondary_index_management | 0 | 1 | 0 | 0
> | 0 | 0
> validation_executor | 0 | 2147483647 | 0 | 0
> | 0 | 0
> view_build_executor | 0 | 1 | 0 | 0
> | 0 | 0
> view_mutation_stage | 0 | 32 | 0 | 0
> | 0 | 0
> {code}
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
---------------------------------------------------------------------
To unsubscribe, e-mail: commits-unsubscr...@cassandra.apache.org
For additional commands, e-mail: commits-h...@cassandra.apache.org
