WamBamBoozle opened a new pull request #31072:
URL: https://github.com/apache/spark/pull/31072
<!--
Thanks for sending a pull request! Here are some tips for you:
1. If this is your first time, please read our contributor guidelines:
https://spark.apache.org/contributing.html
2. Ensure you have added or run the appropriate tests for your PR:
https://spark.apache.org/developer-tools.html
3. If the PR is unfinished, add '[WIP]' in your PR title, e.g.,
'[WIP][SPARK-XXXX] Your PR title ...'.
4. Be sure to keep the PR description updated to reflect all changes.
5. Please write your PR title to summarize what this PR proposes.
6. If possible, provide a concise example to reproduce the issue for a
faster review.
7. If you want to add a new configuration, please read the guideline first
for naming configurations in
'core/src/main/scala/org/apache/spark/internal/config/ConfigEntry.scala'.
-->
### What changes were proposed in this pull request?
If your worker function has a lengthy initialization, and your
application has lots of partitions, you may find you are spending weeks
of compute time repeatedly doing something that should have taken a few
seconds during daemon initialization.
Every Spark executor spawns a process running an R daemon. The daemon
"forks a copy" of itself whenever Spark finds work for it to do. It may
be applying a predefined method such as "max", or it may be applying
your worker function. SparkR::gapply arranges things so that your worker
function will be called with each group. A group is the pair
Key-Seq[Row]. In the absence of partitioning, the daemon will fork for
every group found. With partitioning, the daemon will fork for every
partition found. A partition may have several groups in it.
All the initializations and library loading your worker function manages
is thrown away when the fork concludes. Every fork has to be
initialized.
The configuration spark.r.daemonInit provides a way to avoid reloading
packages every time the daemon forks by having the daemon pre-load
packages. You do this by providing R code to initialize the daemon for
your application.
## Examples
Suppose we want library(wow) to be pre-loaded for our workers.
```R
sparkR.session(spark.r.daemonInit = 'library(wow)')
```
of course, that would only work if we knew that library(wow) was on our
path and available on the executor. If we have to ship the library, we
can use YARN
```R
sparkR.session(
master = 'yarn',
spark.r.daemonInit = '.libPaths(c("wowTarget", .libPaths()));
library(wow)',
spark.submit.deployMode = 'client',
spark.yarn.dist.archives = 'wow.zip#wowTarget')
```
YARN creates a directory for the new executor, unzips 'wow.zip' in some
other directory, and then provides a symlink to it called
./wowTarget. When the executor starts the daemon, the daemon loads
library(wow) from the newly created wowTarget.
Warning: if your initialization takes longer than 10 seconds, consider
increasing the configuration
[spark.r.daemonTimeout](configuration.md#sparkr).
### Why are the changes needed?
<!--
Please clarify why the changes are needed. For instance,
1. If you propose a new API, clarify the use case for a new API.
2. If you fix a bug, you can clarify why it is a bug.
-->
Performance
### Does this PR introduce _any_ user-facing change?
<!--
Note that it means *any* user-facing change including all aspects such as
the documentation fix.
If yes, please clarify the previous behavior and the change this PR proposes
- provide the console output, description and/or an example to show the
behavior difference if possible.
If possible, please also clarify if this is a user-facing change compared to
the released Spark versions or within the unreleased branches such as master.
If no, write 'No'.
-->
Yes. It introduces two new sparkR session parameters as described above.
### How was this patch tested?
<!--
If tests were added, say they were added here. Please make sure to add some
test cases that check the changes thoroughly including negative and positive
cases if possible.
If it was tested in a way different from regular unit tests, please clarify
how you tested step by step, ideally copy and paste-able, so that other
reviewers can test and check, and descendants can verify in the future.
If tests were not added, please describe why they were not added and/or why
it was difficult to add.
-->
Test was added:
[test_daemon_initialization.R](https://github.com/WamBamBoozle/spark/blob/daemon_init/R/pkg/tests/fulltests/test_daemon_initialization.R)
It has been in production for several months now at Target where it has
saved *years* of compute time.
----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
For queries about this service, please contact Infrastructure at:
[email protected]
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
