[
https://issues.apache.org/jira/browse/SPARK-22805?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16293836#comment-16293836
]
Sergei Lebedev edited comment on SPARK-22805 at 12/16/17 11:17 PM:
-------------------------------------------------------------------
I've emulated the effect of SPARK-20923 by removing all
{{"internal.metrics.updatedBlockStatuses"}} entries from the original 79G event
log. The table below compares uncompressed/compressed sizes of this log with
and without the patch proposed in this issue:
||Mode||Size||
|Decompressed|157M|
|Decompressed with patch|155M|
-
*Update*: turns out {{SparkTaskEndEvent}} carries the list of updated blocks
twice (!): as part of the {{"Accumulables"}} and in {{"Task Metrics"}}.
[~andrewor14], [~srowen] do you know if there is a reason for that? It looks
like a bug to me.
*Update*: I'm recomputing the numbers for a fully updatedBlockStatuses-free log.
*Update*: the effect of SPARK-20923 is much more noticeable than I thought
initially. Removing {{"internal.metrics.updatedBlockStatuses"}} from
{{"Accumulables"}} and {{"Updated Blocks"}} from {{"Task Metrics"}} reduced the
log size to 160M. The storage level compression now just shaves of a few M (see
updated table).
was (Author: lebedev):
I've emulated the effect of SPARK-20923 by removing all
{{"internal.metrics.updatedBlockStatuses"}} entries from the original 79G event
log. The table below compares uncompressed/compressed sizes of this log with
and without the patch proposed in this issue:
||Mode||Size||
|LZ4-compressed|2.3G|
|Decompressed|25G|
|LZ4-compressed with patch|2.3G|
|Decompressed with patch|16G|
*Update*: turns out {{SparkTaskEndEvent}} carries the list of updated blocks
twice (!): as part of the {{"Accumulables"}} and in {{"Task Metrics"}}.
[~andrewor14], [~srowen] do you know if there is a reason for that? It looks
like a bug to me.
*Update*: I'm recomputing the numbers for a fully updatedBlockStatuses-free log.
> Use aliases for StorageLevel in event logs
> ------------------------------------------
>
> Key: SPARK-22805
> URL: https://issues.apache.org/jira/browse/SPARK-22805
> Project: Spark
> Issue Type: Improvement
> Components: Spark Core
> Affects Versions: 2.1.2, 2.2.1
> Reporter: Sergei Lebedev
> Priority: Minor
>
> Fact 1: {{StorageLevel}} has a private constructor, therefore a list of
> predefined levels is not extendable (by the users).
> Fact 2: The format of event logs uses redundant representation for storage
> levels
> {code}
> >>> len('{"Use Disk": true, "Use Memory": false, "Deserialized": true,
> >>> "Replication": 1}')
> 79
> >>> len('DISK_ONLY')
> 9
> {code}
> Fact 3: This leads to excessive log sizes for workloads with lots of
> partitions, because every partition would have the storage level field which
> is 60-70 bytes more than it should be.
> Suggested quick win: use the names of the predefined levels to identify them
> in the event log.
--
This message was sent by Atlassian JIRA
(v6.4.14#64029)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
