[
https://issues.apache.org/jira/browse/CASSANDRA-18283?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17694536#comment-17694536
]
Brad Schoening commented on CASSANDRA-18283:
--------------------------------------------
[~smiklosovic] I've added two examples above in the description. Below an
example we're seeing in prod with TWCS where something is clearly awry. This
output is from instaclustr's sstablemetadata script. For nodetool tablestats,
showing the max duration would be sufficient to know that something is not
working right.
compaction =
\{'class':'org.apache.cassandra.db.compaction.TimeWindowCompactionStrategy','compaction_window_size':'15','compaction_window_unit':'DAYS','
...}
!image-2023-02-28-08-08-24-727.png!
the current nodetool tablestats for this table is rather basic and doesn't
suggest a problem.
Table: hourly_series
SSTable count: 47
Space used (live): 322865109897
Space used (total): 322865109897
> Enhance nodetool tablestats
> ---------------------------
>
> Key: CASSANDRA-18283
> URL: https://issues.apache.org/jira/browse/CASSANDRA-18283
> Project: Cassandra
> Issue Type: Improvement
> Components: Tool/nodetool
> Reporter: Brad Schoening
> Assignee: Stefan Miklosovic
> Priority: Normal
> Fix For: 4.x
>
> Attachments: image-2023-02-28-08-08-24-727.png
>
> Time Spent: 10m
> Remaining Estimate: 0h
>
> The nodetool tablestats command lacks some available details which would be
> very useful to report upon. This is especially helpful in
> database-as-a-service environments where servers and their disk files are not
> directly observable by users.
> 1. Currently, for LCS tablestats reports useful details about the number of
> sstables in each level:
> SSTable count: 6635
> SSTables in each level: [1, 9, 98, 805, 5722, 0, 0, 0, 0]
> This type of additional detail about the sstables is absent from STCS and
> TWCS as it only reports the table count.
> 1a) For STCS, tablestats should report the max sstable file size on disk.
> This is useful to know if compaction has failed due to disk space or if a
> forced compaction created a jumbo table.
> 1b) For TWCS, tablestats should report the min & max timestamp, and duration
> of the sstables representing windows. This is useful to know if
> out-of-window writes or rows w/out a TTL have lead many more sstables on disk
> than expected by the time window configuration.
> STCs example:
> SSTable count: 6635
> SSTable STCS max size: 122,000,000,000
> STCs example:
> SSTable count: 6635
> SSTables Time Window 15 DAYS, max duration : 362d 7h 16m 49s
> 2. While tablestats reports both memtable and disk file sstable statistics.
> It is useful these are in the same command, but it would clarify the output
> to separate mem vs disk into two sections
> i.e.,
> -- File statistics
> SSTable count: 6635
> SSTables in each level: [1, 9, 98, 805, 5722, 0, 0, 0, 0]
> -- Memtable statistics
> Bloom filter false positives: 12184123
> Bloom filter false ratio: 0.07203
> Bloom filter space used: 16874424
> Bloom filter off heap memory used: 16821344
> Index summary off heap memory used: 7525546
> Space used (live): 1324067896238
> 3. Read / Write count should also be reported as a ratio, such as:
> Local read count: 202961459
> Local write count: 40554481
> Local read/write ratio: 5:1 <new>
> Local read latency: 1.957 ms
> Local write count: 40554481
> Local write latency: 0.040 ms
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
