[SPARK-26082] Misnaming of spark.mesos.fetch(er)Cache.enable in MesosClusterScheduler - ASF JIRA

Rank to Top

Rank to Bottom

Attach files

Attach Screenshot

Bulk Copy Attachments

Bulk Move Attachments

Voters

Watch issue

Watchers

Create sub-task

Convert to sub-task

Link

Clone

Labels

Update Comment Author

Replace String in Comment

Update Comment Visibility

Delete Comments

XML

Word

Printable

JSON

Details

Type: Bug
Status: Resolved
Priority: Major
Resolution: Fixed
Affects Version/s: 2.1.0, 2.1.1, 2.1.2, 2.1.3, 2.2.0, 2.2.1, 2.2.2, 2.3.0, 2.3.1, 2.3.2
Fix Version/s: 2.3.4, 2.4.1, 3.0.0
Component/s: Mesos
Labels:
None

Description

Currently in docs:

spark.mesos.fetcherCache.enable / false / If set to `true`, all URIs (example: `spark.executor.uri`, `spark.mesos.uris`) will be cached by the Mesos Fetcher Cache

Currently in MesosClusterScheduler.scala (which passes parameter to driver):
private val useFetchCache = conf.getBoolean("spark.mesos.fetchCache.enable", false)

Currently in MesosCourseGrainedSchedulerBackend.scala (which passes mesos caching parameter to executors):
private val useFetcherCache = conf.getBoolean("spark.mesos.fetcherCache.enable", false)

This naming discrepancy dates back to version 2.0.0 (jira).

This means that when spark.mesos.fetcherCache.enable=true is specified, the Mesos cache will be used only for executors, and not for drivers.

IMPACT:
Not caching these driver files (typically including at least spark binaries, custom jar, and additional dependencies) adds considerable overhead network traffic and startup time when frequently running spark Applications on a Mesos cluster. Additionally, since extracted files like spark-x.x.x-bin-*.tgz are additionally copied and left in the sandbox with the cache off (rather than extracted directly without an extra copy), this can considerably increase disk usage. Users CAN currently workaround by specifying the spark.mesos.fetchCache.enable option, but this should at least be specified in the documentation.

SUGGESTED FIX:
Add spark.mesos.fetchCache.enable to the documentation for versions 2 - 2.4, and update MesosClusterScheduler.scala to use spark.mesos.fetcherCache.enable going forward (literally a one-line change).