We wish to use the latest build of presto, which includes an update of the bundled version of Aluxio to 2.9.3.
See: https://github.com/prestodb/presto/pull/19594
We believe that this will fix an issue using the local caching feature of Presto with HDFS external storage.
| Status | Subtype | Assigned | Task | ||
|---|---|---|---|---|---|
| Open | None | T288252 Alluxio for Improved Superset Query Performance | |||
| Open | None | T266641 [Data Platform] Test Alluxio as cache layer for Presto | |||
| Resolved | BTullis | T342343 Upgrade Presto to version 0.283 |
Presto version 0.284 was released yesterday: https://github.com/prestodb/presto/releases/tag/0.284
The release notes seem to have been slightly delayed, but are available here.
There are some changes mentioned regarding cache performance for the history based optimizer, plus some hive changes and some iceberg changes.
However, now that I look again, it seems that the upgrade of the alluxio client jar was reverted.
Here is the pom.xml file at version 0.284.
It shows that the Alluxio version is back at 2.8.1
It was reverted in this commit, which was between version 0.283 and 0.284. I wonder why.
So for now I suggest that we don't upgrade beyond 0.283 and we try to find out why the alluxio jar was downgraded.
Beginning work on the new build now, referring to the build for 0.281 T337335 and the README.Debian file.
Configured the version:
export VERSION=0.283
Downloaded the release tarball:
wget -P /tmp https://repo1.maven.org/maven2/com/facebook/presto/presto-server/$VERSION/presto-server-$VERSION.tar.gz
Checked the checksum
[ "$(curl -s https://repo1.maven.org/maven2/com/facebook/presto/presto-server/$VERSION/presto-server-$VERSION.tar.gz.sha1)" == "$(shasum -a 1 /tmp/presto-server-$VERSION.tar.gz | awk '{print $1}')" ] && echo "presto-server sha1 checksum matches, continue" || echo "presto-server sha1 checksum does not match!"The following output was shown.
presto-server sha1 checksum matches, continue
Updated the tarball from which the debian package is built.
gbp import-orig -u $VERSION --upstream-branch=master --debian-branch=debian --merge-mode=replace /tmp/presto-server-$VERSION.tar.gz
The following output was shown:
gbp:info: Importing '/tmp/presto-server-0.283.tar.gz' to branch 'master'... gbp:info: Source package is presto gbp:info: Upstream version is 0.283 gbp:info: Replacing upstream source on 'debian' gbp:info: Successfully imported version 0.283 of /tmp/presto-server-0.283.tar.gz
Removed the presto-cli jar file.
git rm debian/lib/presto-cli-*-executable.jar
The following output was shown:
rm 'debian/lib/presto-cli-0.281-executable.jar'
Downloaded the new jar
wget -P debian/lib https://repo1.maven.org/maven2/com/facebook/presto/presto-cli/$VERSION/presto-cli-$VERSION-executable.jar
Check the checksum of this jar:
[ "$(curl -s https://repo1.maven.org/maven2/com/facebook/presto/presto-cli/$VERSION/presto-cli-$VERSION-executable.jar.sha1)" == "$(shasum -a 1 ./debian/lib/presto-cli-$VERSION-executable.jar | awk '{print $1}')" ] && echo "presto-cli sha1 checksum matches, continue" || echo "presto-cli sha1 checksum does not match!"The following output was shown:
presto-cli sha1 checksum matches, continue
Configured the presto-cli jar to be executable:
chmod 755 debian/lib/presto-cli-$VERSION-executable.jar
Find the binaries that need to be included:
find {debian/lib,lib,plugin} -type f -exec file {} \; | grep -v text | awk -F ':' '{print $1}' | sort > debian/source/include-binariesEdited the version of the presto-cli jar file mentioned in in debian/presto-cli.links and debian/presto-cli.install from 0.281 to 0.283
Created a new changelog entry with:
dch -v 0.283-1 -D bullseye-wikimedia --force-distribution
Commited with:
git add debian git commit -m "Upstream release $VERSION"
I pushed both the master and debian branches to gerrit. Last time I forgot to push the master branch, which resulted in a build failure.
(base) btullis@marlin:~/wmf/debs/presto$ git push Enumerating objects: 261, done. Counting objects: 100% (220/220), done. Delta compression using up to 16 threads Compressing objects: 100% (170/170), done. Writing objects: 100% (170/170), 139.65 MiB | 3.10 MiB/s, done. Total 170 (delta 46), reused 0 (delta 0), pack-reused 0 remote: Resolving deltas: 100% (46/46) remote: Processing changes: refs: 1, done To ssh://gerrit.wikimedia.org:29418/operations/debs/presto d2f7f77..d45707c debian -> debian (base) btullis@marlin:~/wmf/debs/presto$ git checkout master Switched to branch 'master' Your branch is ahead of 'origin/master' by 1 commit. (use "git push" to publish your local commits) (base) btullis@marlin:~/wmf/debs/presto$ git push Total 0 (delta 0), reused 0 (delta 0), pack-reused 0 remote: Processing changes: refs: 1, done To ssh://gerrit.wikimedia.org:29418/operations/debs/presto 3ecfb7e..af0a3de master -> master
On build2001 - check out the repository.
git clone "https://gerrit.wikimedia.org/r/operations/debs/presto" && (cd "presto" && mkdir -p `git rev-parse --git-dir`/hooks/ && curl -Lo `git rev-parse --git-dir`/hooks/commit-msg https://gerrit.wikimedia.org/r/tools/hooks/commit-msg; chmod +x `git rev-parse --git-dir`/hooks/commit-msg)
Check that the version on the debian branch is correct.
btullis@build2001:~$ cd presto/
btullis@build2001:~/presto$ git checkout debian
Branch 'debian' set up to track remote branch 'debian' from 'origin'.
Switched to a new branch 'debian'
btullis@build2001:~/presto$ git log -n 1
commit d45707cce2a8c87070384aea64d98044842985ec (HEAD -> debian, origin/debian)
Author: Ben Tullis <btullis@wikimedia.org>
Date: Tue Oct 10 14:53:16 2023 +0100
Upstream release 0.283
Change-Id: I1a2a9bf1e2ec8b19b63d1918aad2e05d862bed08
btullis@build2001:~/presto$Now issue the package build command.
GIT_PBUILDER_AUTOCONF=no DIST=bullseye WIKIMEDIA=yes gbp buildpackage -sa -us -uc --git-builder=git-pbuilder --source-option="--include-removal"
Copied the debs to apt1001.
btullis@apt1001:~$ rsync rsync://build2001.codfw.wmnet/pbuilder-result/bullseye-amd64/presto* .
Added the debs to the apt-repository with:
btullis@apt1001:~$ sudo -i reprepro include bullseye-wikimedia `pwd`/presto_0.283-1_amd64.changes btullis@apt1001:~$ sudo -i reprepro --ignore=wrongdistribution include buster-wikimedia `pwd`/presto_0.283-1_amd64.changes
Checked that they are available for installation.
btullis@apt1001:~$ sudo -i reprepro ls presto-cli presto-cli | 0.283-1 | buster-wikimedia | amd64, i386 presto-cli | 0.283-1 | bullseye-wikimedia | amd64, i386 btullis@apt1001:~$ sudo -i reprepro ls presto-server presto-server | 0.283-1 | buster-wikimedia | amd64, i386 presto-server | 0.283-1 | bullseye-wikimedia | amd64, i386
I'm pushing out the new version of presto to the test cluster with:
btullis@cumin1001:~$ sudo debdeploy deploy -u 2023-10-12-presto.yaml -Q 'P{O:analytics_test_cluster::presto::server} or P{O:analytics_test_cluster::coordinator}'I've done some basic tests on the test cluster and things appear to be working properly.
presto> SELECT node_id,node_version FROM system.runtime.nodes;
node_id | node_version
--------------------------------+---------------
an-test-coord1001-eqiad-wmnet | 0.283-1fa586a
an-test-presto1001-eqiad-wmnet | 0.283-1fa586a
(2 rows)
Query 20231012_131619_00009_h599v, FINISHED, 2 nodes
Splits: 17 total, 17 done (100.00%)
[Latency: client-side: 219ms, server-side: 182ms] [2 rows, 184B] [10 rows/s, 1010B/s]
presto> select uri_host,uri_path from wmf.webrequest where year=2023 and month=10 and day=12 and hour=4 limit 10;
uri_host | uri_path
--------------------+-------------------------
zh.wikipedia.org | /w/api.php
en.m.wikipedia.org | /wiki/Beer_on_the_Table
meta.wikimedia.org | /w/index.php
en.wikipedia.org | /w/api.php
en.wikipedia.org | /wiki/John_Fetterman
de.wikipedia.org | /w/api.php
www.wikipedia.org | /
en.m.wikipedia.org | /wiki/Greater_Khorasan
en.wikipedia.org | /w/load.php
meta.wikimedia.org | /w/index.php
(10 rows)
Query 20231012_131637_00010_h599v, FINISHED, 1 node
Splits: 273 total, 59 done (21.61%)
[Latency: client-side: 0:01, server-side: 0:01] [630 rows, 721KB] [914 rows/s, 1.02MB/s]@JAllemandou - should I push this out to the production presto cluster today, or would you rather that we do some more testing on this cluster first?
Hi @BTullis - I'm sorry I missed the ping yesterday.
I think we can go to prod with the new version.
Let's do that on Monday :)
Mentioned in SAL (#wikimedia-analytics) [2023-10-16T10:06:03Z] <btullis> deploying presto version 0.283 to production for T342343 with sudo debdeploy deploy -u 2023-10-12-presto.yaml -Q 'P{O:analytics_cluster::presto::server} or P{O:analytics_cluster::coordinator} or A:stat'
This looks good. All presto servers and client deployed.
btullis@stat1009:~$ presto --catalog analytics_hive
Picked up JAVA_TOOL_OPTIONS: -Dfile.encoding=UTF-8
presto> SELECT node_id,node_version FROM system.runtime.nodes;
node_id | node_version
---------------------------+---------------
an-coord1001-eqiad-wmnet | 0.283-1fa586a
an-presto1001-eqiad-wmnet | 0.283-1fa586a
an-presto1002-eqiad-wmnet | 0.283-1fa586a
an-presto1003-eqiad-wmnet | 0.283-1fa586a
an-presto1004-eqiad-wmnet | 0.283-1fa586a
an-presto1005-eqiad-wmnet | 0.283-1fa586a
an-presto1006-eqiad-wmnet | 0.283-1fa586a
an-presto1007-eqiad-wmnet | 0.283-1fa586a
an-presto1008-eqiad-wmnet | 0.283-1fa586a
an-presto1009-eqiad-wmnet | 0.283-1fa586a
an-presto1010-eqiad-wmnet | 0.283-1fa586a
an-presto1011-eqiad-wmnet | 0.283-1fa586a
an-presto1012-eqiad-wmnet | 0.283-1fa586a
an-presto1013-eqiad-wmnet | 0.283-1fa586a
an-presto1014-eqiad-wmnet | 0.283-1fa586a
an-presto1015-eqiad-wmnet | 0.283-1fa586a
(16 rows)
Query 20231016_101803_00018_vrxm9, FINISHED, 2 nodes
Splits: 17 total, 17 done (100.00%)
[Latency: client-side: 344ms, server-side: 290ms] [16 rows, 1.29KB] [55 rows/s, 4.44KB/s]
presto>A simple query runs successfully on 15 nodes.
presto> select uri_host,uri_path from wmf.webrequest where year=2023 and month=10 and day=16 and hour=4 limit 10;
uri_host | uri_path
--------------------+----------------------------------------------------------------------------------------------
en.wikipedia.org | /wiki/File:World_energy_consumption.svg
en.m.wikipedia.org | /w/api.php
www.wikidata.org | /wiki/Special:EntityData/Q117789109.ttl
es.m.wikipedia.org | /w/load.php
el.wikipedia.org | /api/rest_v1/page/summary/%CE%92%CE%B1%CF%80%CF%84%CE%B9%CF%83%CF%84%CE%AE%CF%81%CE%B9%CE%BF
en.wikipedia.org | /w/load.php
fa.m.wikipedia.org | /w/load.php
en.wikipedia.org | /w/load.php
en.wikipedia.org | /w/extensions/Wikibase/client/resources/images/edit.svg
it.m.wikipedia.org | /w/load.php
(10 rows)
Query 20231016_101923_00019_vrxm9, FINISHED, 15 nodes
Splits: 1,391 total, 810 done (58.23%)
[Latency: client-side: 0:04, server-side: 0:04] [2.81K rows, 1.3GB] [655 rows/s, 309MB/s]
presto>