[
https://issues.apache.org/jira/browse/CASSANDRA-19015?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17786428#comment-17786428
]
Brad Schoening edited comment on CASSANDRA-19015 at 11/15/23 6:37 PM:
----------------------------------------------------------------------
The current output is not just too many decimals, it is also an inconsistent
number of decimal places. The result is neither human nor machine friendly.
For example,
{noformat}
Read Latency: 8.894851803649942 ms
Local read latency: 6.857 ms
{noformat}
What should be corrected here is:
* *inconsistent display and rounding* for the same units of measure and same
scale (i.e., latency)
* *accuracy* - reporting milliseconds to 15 decimal places is inaccurate
because (most) CPUs can measure beyond a few nanoseconds.
The display length issues are due to IEEE floating point representation. I can
see the StatsTable.java uses the double type for latency in MS.
{code:java}
StatsKeyspace:
private double totalReadTime;
private double totalWriteTime;
StatsTable.java
public double localReadLatencyMs;
public double localWriteLatencyMs; {code}
but inconsistently applies a *%01.3f* format to it:
{code:java}
TableStatsPrinter.java:
missing format on these doubles in ms:
out.println("\tRead Latency: " + keyspace.readLatency() + " ms");
out.println("\tWrite Latency: " + keyspace.writeLatency() + " ms");
but correctly formats these doubles in ms:
out.printf(indent + "Local read latency: %01.3f ms%n",
table.localReadLatencyMs);
out.printf(indent + "Local write latency: %01.3f ms%n",
table.localWriteLatencyMs);{code}
We should maybe leave the thousands separator for a different Jira and use this
Jira to focus on displaying floating point numbers with an appropriate and
consistent number of decimal places. I see now there is a -H option to display
bytes in KiB, MiB, etc. In addition there is also a -F format option to display
results in a machine readable json or yaml.
was (Author: bschoeni):
The current output is not just too many decimals, it is also an inconsistent
number of decimal places. The result is neither human nor machine friendly.
For example,
{noformat}
Read Latency: 8.894851803649942 ms
Local read latency: 6.857 ms
{noformat}
What should be corrected here is:
* *inconsistent display and rounding* for the same units of measure and same
scale (i.e., latency)
* *accuracy* - reporting milliseconds to 15 decimal places is inaccurate
because (most) CPUs can measure beyond a few nanoseconds.
The display length issues are due to IEEE floating point representation. I can
see the StatsTable.java uses the double type for latency in MS.
{code:java}
StatsKeyspace:
private double totalReadTime;
private double totalWriteTime;
StatsTable.java
public double localReadLatencyMs;
public double localWriteLatencyMs; {code}
but inconsistently applies a format to it:
{code:java}
TableStatsPrinter.java:
missing format on these doubles in ms:
out.println("\tRead Latency: " + keyspace.readLatency() + " ms");
out.println("\tWrite Latency: " + keyspace.writeLatency() + " ms");
but correctly formats these doubles in ms:
out.printf(indent + "Local read latency: %01.3f ms%n",
table.localReadLatencyMs);
out.printf(indent + "Local write latency: %01.3f ms%n",
table.localWriteLatencyMs);{code}
We should maybe leave the thousands separator for a different Jira and use this
Jira to focus on displaying floating point numbers with an appropriate and
consistent number of decimal places. I see now there is a -H option to display
bytes in KiB, MiB, etc. In addition there is also a -F format option to display
results in a machine readable json or yaml.
> Nodetool 'tablestats' formatting uses inconsistent significant digits
> ---------------------------------------------------------------------
>
> Key: CASSANDRA-19015
> URL: https://issues.apache.org/jira/browse/CASSANDRA-19015
> Project: Cassandra
> Issue Type: Improvement
> Components: Tool/nodetool
> Reporter: Brad Schoening
> Assignee: Leo
> Priority: Low
>
> Nodetool reports milliseconds (ms) with anywhere from 3 to 15 significant
> digits. Ratios use five or sixteen decimal places. Averages use 1 or 13
> decimal places.
> * milliseconds should use 3 decimal places
> * ratios should use 3 decimal places (tenths of a percent)
> * averages should use 1 or 2
> For readability, it would be helpful if large integers had comma separators.
> I.e., space used: as 1,463,210,998,523 and/or in GiB/MiB/KiB. It's unclear
> if the exact disk size is somehow useful, as it may change minute-by-minute,
> if not, rounding would be best, or displaying both Space used (live):
> 1,463,210,998,523 (1,463GiB)
> Total number of tables: 83
> ----------------
> Keyspace : X
> Read Count: 1007337271
> Read Latency: 8.485891803649942 ms
> Write Count: 67550181
> Write Latency: 0.02556443163342523 ms
> Pending Flushes: 0
> Table: Y
> SSTable count: 7183
> Old SSTable count: 0
> SSTables in each level: [0, 9, 92, 754, 6328, 0, 0, 0, 0]
> Space used (live): 1463210998523
> Space used (total): 1463210998523
> Space used by snapshots (total): 0
> Off heap memory used (total): 607419608
> SSTable Compression Ratio: 0.3146620992793412
> Number of partitions (estimate): 24784137
> Memtable cell count: 106067
> Memtable data size: 248539982
> Memtable off heap memory used: 0
> Memtable switch count: 256
> Local read count: 865440924
> Local read latency: 6.857 ms
> Local write count: 13881409
> Local write latency: 0.037 ms
> Pending flushes: 0
> Percent repaired: 0.0
> Bytes repaired: 0.000KiB
> Bytes unrepaired: 4315.386GiB
> Bytes pending repair: 0.000KiB
> Bloom filter false positives: 11027855
> Bloom filter false ratio: 0.01099
> Bloom filter space used: 33590024
> Bloom filter off heap memory used: 33532560
> Index summary off heap memory used: 8174024
> Compression metadata off heap memory used: 565713024
> Compacted partition minimum bytes: 36
> Compacted partition maximum bytes: 17797419593
> Compacted partition mean bytes: 189740
> Average live cells per slice (last five minutes): 1443.2146104466253
> Maximum live cells per slice (last five minutes): 105778
> Average tombstones per slice (last five minutes): 1.0
> Maximum tombstones per slice (last five minutes): 1
> Dropped Mutations: 0
> Droppable tombstone ratio: 0.00000
--
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
