[
https://issues.apache.org/jira/browse/CASSANDRA-19776?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17955920#comment-17955920
]
Branimir Lambov commented on CASSANDRA-19776:
---------------------------------------------
Yes, (1) will be false unless we use bigger input sstables (at least 10s of
MiB) together with a smaller preemptive open interval (e.g. 2MiB).
As I understand it the problem is caused by that early checkpointing. A similar
thing also happens when we switch to a new output sstable in the same
operation, again when the size of the output is large enough for that to
happen; if you want to go that route, make sure the inputs are again large
enough, that the base shard count is at least 2, and that the minumum sstable
size is set small (e.g. 2MiB).
> Spinning trying to capture readers
> ----------------------------------
>
> Key: CASSANDRA-19776
> URL: https://issues.apache.org/jira/browse/CASSANDRA-19776
> Project: Apache Cassandra
> Issue Type: Bug
> Components: Legacy/Core
> Reporter: Cameron Zemek
> Assignee: Stefan Miklosovic
> Priority: Normal
> Fix For: 4.0.x, 4.1.x, 5.0.x, 5.x
>
> Attachments: extract.log
>
> Time Spent: 50m
> Remaining Estimate: 0h
>
> On a handful of clusters we are noticing Spin locks occurring. I traced back
> all the calls to the EstimatedPartitionCount metric (eg.
> org.apache.cassandra.metrics:type=Table,keyspace=testks,scope=testcf,name=EstimatedPartitionCount)
> Using the following patched function:
> {code:java}
> public RefViewFragment selectAndReference(Function<View,
> Iterable<SSTableReader>> filter)
> {
> long failingSince = -1L;
> boolean first = true;
> while (true)
> {
> ViewFragment view = select(filter);
> Refs<SSTableReader> refs = Refs.tryRef(view.sstables);
> if (refs != null)
> return new RefViewFragment(view.sstables, view.memtables,
> refs);
> if (failingSince <= 0)
> {
> failingSince = System.nanoTime();
> }
> else if (System.nanoTime() - failingSince >
> TimeUnit.MILLISECONDS.toNanos(100))
> {
> List<SSTableReader> released = new ArrayList<>();
> for (SSTableReader reader : view.sstables)
> if (reader.selfRef().globalCount() == 0)
> released.add(reader);
> NoSpamLogger.log(logger, NoSpamLogger.Level.WARN, 1,
> TimeUnit.SECONDS,
> "Spinning trying to capture readers {},
> released: {}, ", view.sstables, released);
> if (first)
> {
> first = false;
> try {
> throw new RuntimeException("Spinning trying to
> capture readers");
> } catch (Exception e) {
> logger.warn("Spin lock stacktrace", e);
> }
> }
> failingSince = System.nanoTime();
> }
> }
> }
> {code}
> Digging into this code I found it will fail if any of the sstables are in
> released state (ie. reader.selfRef().globalCount() == 0).
> See the extract.log for an example of one of these spin lock occurrences.
> Sometimes these spin locks last over 5 minutes. Across the worst cluster with
> this issue, I ran a log processing script that everytime the 'Spinning trying
> to capture readers' was different to previous one it would output if the
> released tables were in Compacting state. Every single occurrence has it spin
> locking with released listing a sstable that is compacting.
> In the extract.log example its spin locking saying that nb-320533-big-Data.db
> has been released. But you can see prior to it spinning that sstable is
> involved in a compaction. The compaction completes at 01:03:36 and the
> spinning stops. nb-320533-big-Data.db is deleted at 01:03:49 along with the
> other 9 sstables involved in the compaction.
>
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: commits-unsubscr...@cassandra.apache.org
For additional commands, e-mail: commits-h...@cassandra.apache.org
