Parsoid

Parsoid is a service that converts between wikitext and HTML. The HTML contains additional metadata that allows it to be converted back ("round-tripped") to wikitext. Parsoid operates as a stateless HTTP server running on port 8000.

Uses

• VisualEditor fetches the HTML for a given page from Parsoid, edits it, then delivers the modified HTML to Parsoid, which converts it back to wikitext.
• Flow (as configured on WMF wikis with $wgFlowContentFormat = 'html') works the other way around. When a user creates a post Flow uses Parsoid to convert the wikitext to HTML and Flow stores the HTML in ExternalStore. If someone later edits a post Flow uses Parsoid to convert the HTML back to wikitext for editing.

Monitoring

• Parsoid eqiad cluster in Grafana
• Parsoid codfw cluster in Grafana
• Icinga has service checks for HTTP on port 8000 on both the individual backends and on the LVS service IP.
• pybal does health checks on all backends every second, and depools boxes that are down as long as the % of depooled boxes does not exceed 50%. To see these health checks and depools/repools happen in real time, run ssh parsoid.svc.eqiad.wmnet (this will drop you into either lvs1003 or lvs1006, depending on which is active), then tail -f /var/log/pybal.log | grep parsoid
• Logging happens in /var/log/parsoid/parsoid.log. There is a log rotation setup in /etc/logrotate.d/parsoid.
• Logs in logstash (Parsoid/PHP): https://logstash.wikimedia.org/app/dashboards#/view/AW4Y6bumP44edBvO7lRc
• Logging in Parsoid starts at "warn" level (see https://github.com/wikimedia/operations-mediawiki-config/blob/master/wmf-config/InitialiseSettings.php#L5459)
• Useful links for Parsoid deployers
• Currently running Parsoid version:
  • In beta: https://en.wikipedia.beta.wmflabs.org/wiki/Special:Version#mw-version-library-wikimedia/parsoid
  • In production: https://en.wikipedia.org/wiki/Special:Version#mw-version-library-wikimedia/parsoid
  • On scandium: See https://www.mediawiki.org/wiki/Parsoid/Round-trip_testing
  • (Be aware that the parsoid cluster is behind restbase and although the cluster *should* be running the same version of Parsoid as the mediawiki frontends, if puppet or scap are broken (esp in beta) things could diverge.)

Machine overview

These are the machines involved in a Parsoid deploy:

• In the beta/wmflabs cluster:
  • deployment-deploy01.deployment-prep.eqiad.wmflabs: staging host in beta; no longer used.
  • deployment-parsoid11.deployment-prep.eqiad.wmflabs: parsoid server in beta
  • deployment-restbase02.deployment-prep.eqiad.wmflabs: restbase server in beta
• In the production cluster:
  • deployment.eqiad.wmnet: staging host in production; no longer used
  • wtp1xxx: parsoid servers in eqiad cluster
  • restbase1xxx: restbase servers in eqiad cluster
  • parse2xxx: parsoid servers in codfw cluster
  • restbase2xxx: restbase servers in codfw cluster
  • scandium.eqiad.wmnet: Parsoid testing host, has read-only access to the production database.

Deploying changes

Parsoid is deployed as part of the MediaWiki train. See How to deploy code for an overview, Heterogeneous deployment for a more technical description of the directory structures involved, and Heterogeneous deployment/Train deploys for the steps to do a train deploy. When code changes outside the train schedule are required, a Backport windows will be required. Generally Parsing team members won't be doing train deploys or Backport deploys directly; we will tag a Parsoid version (which releases it to packagist to make it available via composer) and merge a version bump into the mediawiki/vendor repository. Once the patch is merged into vendor, the new version of Parsoid goes live in beta (almost) immediately; it will then be rolled out to production on the next train.

Deploying Parsoid

Test the version you hope to deploy

• See mw:Parsoid/Round-trip testing for details.
• Check http://parsoid-rt-tests.wikimedia.org/regressions/between/{from}/{to} where {from} is the last deployed hash from mw:Parsoid/Deployments and {to} is the latest tested commit (which we're about to deploy)
  • http://parsoid-rt-tests.wikimedia.org/commits gives you a nice radio-button interface to create this URL
  • BEWARE: if you get the output total regressions between selected revisions: 0, it is extremely likely that you mistyped the hash or that we didn't actually run round-trip tests for that particular hash. (This is a bug, we should probably give a better message in this case.)
  • Since we are using current revision of titles in round-trip testing, edits to pages can show up as false regressions. tools/regression-testing.php in the Parsoid repo is useful in filtering those out. Running it with the right parameters (use --help for usage) will get a list of pages to look more closely, if necessary.
• Check that there are no concerning notices or errors in logstash from the rt run
  • https://logstash.wikimedia.org/app/dashboards#/view/parsoid-tests

Prepare the vendor patch

Here is a concise summary of steps in the common case. Detailed explanation follows.

cd PARSOID_REPO
git checkout <git-sha-of-patch-to-tag>
git tag v0.{version}.0-a{N}
git push origin v0.{version}.0-a{N}
cd VENDOR_REPO
.. edit composer.json and bump version number of wikimedia/parsoid as above ..
composer update --no-dev
.. ensure all files are added and git commit (see below for what to include in commit message) ..
git review -u
.. add reviewers and get it reviewed ..
.. post-merge, verify it landed on the beta cluster and works fine ..

Details

(This process was hashed out in phab:T240055)

• Pull the latest version of master into your master branch of Parsoid and do remote update thereafter
• Tag a new version of Parsoid and push the tag: (hint use: git tag -l to show existing tags)
  • git tag v0.16.0-a{N}
  • git push origin v0.16.0-a{N} (Include the leading 'v', and substitute the next version number for {N}.)
    • The "origin" remote here is ssh://USER@gerrit.wikimedia.org:29418/mediawiki/services/parsoid
    • Nothing more than usual push permissions on the parsoid repo should be needed. If you need to tweak permissions (for example, to temporarily add force push permissions to fix a mistake), you can do this using the "edit" button at https://gerrit.wikimedia.org/r/#/admin/projects/mediawiki/services/parsoid,access
  • Check that this version has been picked up at https://packagist.org/packages/wikimedia/parsoid (might take a minute, you can work on the deployment summary while you wait)
• Create a short deployment summary on mw:Parsoid/Deployments.
  • In Parsoid repository, tools/gen_deploy_log.sh v0.16.0-a{from} v0.16.0-a{to} (for appropriate values of {from} and {to}) will generate wikitext you can cut-and-paste into mw:Parsoid/Deployments (improvements to this script are welcome!)
  • In mw:Parsoid/Deployments, copy previous release header line, edit the dates and version info and delete "done" template and insert "In progress" template
  • The manual way is/was to start from git log --cherry-pick {from}...{to}. Don't include all commits, but only notable fixes and changes (ignore rt-test fixes, code cleanup updates, parser test updates, etc). (The above command will do the right thing if {from} was on a branch and had patches cherry-picked from {to}, although if there were conflicts during the cherry-pick to {from} the patch will still appear in the log for {to}.)
• Checkout mediawiki/vendor.git master branch into its own working directory. (hint: $ git clone "https://gerrit.wikimedia.org/r/mediawiki/vendor")
  • Make a new branch in that repo: (hint: git branch deploy; get checkout deploy)
  • In that repo: Update composer.json to include "wikimedia/parsoid": "0.16.0-aN", (for your version {N}; note no leading "v")
  • Ensure you're running the version of composer listed in the README for the vendor repo. At time of writing this is 2.2.4. composer --version will tell you what version you're running and (usually) composer self-update will bring you up-to-date.
  • Ensure that you are using the latest version of composer (using composer self-update). Informally, you need to be using "the same version JamesF is using." If you use an old composer, you will create unrelated diffs to non-parsoid code when you do the next step.
  • Do composer update --no-dev (which should only update parsoid)
    • If composer complains "The requested package wikimedia/parsoid 0.16.0-aN exists as [...long list not including 0.16.0-aN...]" then composer's local cache hasn't been updated to include the new version available from packagist.org yet. Wait 15 minutes and try again. The --no-cache option to composer *might* help... but it might not (it probably won't). Apparently composer 2.x sped this up? :)
  • Add the changed files to git, commit and provide a detailed commit message as described below, and then upload to gerrit:

git add wikimedia/parsoid composer.lock composer.json composer # & etc, if needed
git commit
git review

• Use a commit message that (1) names the new parsoid tag, (2) includes the git hash of the new parsoid version (we've stopped including this for the most part because the hash is given by the parsoid tag in part 1), and (3) references key bug #s from the deployment summary so the deploy gets linked to phab ( Tip: git log v0.16.0-a$PREV..v0.16.0-a$NEW | grep Bug: | sort -u). For example:

Bump parsoid to 0.16.0-aN

This corresponds to Parsoid commit cafecafecafecafe.

Bug: T111111
Bug: T222222

• Review the generated patch (either via git show or on gerrit), looking specifically for unexpected changes. The code in wikimedia/parsoid should change in roughly the ways you expect from the deploy summary, there should be a change to the version number in composer.json and changes to some hashes, timestamps, and versions in composer.lock and composer/installed.json, but there should be no other changes. See this patch set for an example where an old version of composer was used, resulting in spurious changes to other files in composer/.
• If jenkins fails on gerrit with the same "The requested package wikimedia/parsoid 0.16.0-aN exists as..." message described above, the reason is the as that described for the composer update --no-dev step above: composer's cache on jenkins still doesn't have your new version yet. Wait a minute and comment "recheck" to re-run the jenkins tests.
• Review and C+2 on gerrit. This will go live on beta cluster pretty quickly (within 30 minutes).
• If you were late and just missed the train branch, be sure to check the "If the train branch has already been cut" section below.

Verify deployment version on beta after the vendor patch is merged

• Check that this is live on the mediawiki front ends in beta by watching the version number listed on https://en.wikipedia.beta.wmflabs.org/wiki/Special:Version#mw-version-library-wikimedia/parsoid
• If ever you need to, you can check the version on parsoid11 as well:

$ ssh deployment-parsoid11.deployment-prep.eqiad.wmflabs
user@deployment-parsoid11$ curl -x deployment-parsoid11:80 'http://en.wikipedia.beta.wmflabs.org/wiki/Special:Version' | fgrep wikimedia/parsoid -C0

• If beta cluster is down or visual editor is down in beta cluster, do not continue with routine deployments.
• On beta cluster (eg en.wikipedia.beta.wmflabs.org), perform manual VisualEditor editing tests. This requires you to have an account on the beta cluster wiki. Test with non-ASCII content too to catch encoding issues. Check parsoid logs. Be particularly alert to integration issues: library conflicts, etc.
• Watch the logs on beta: https://wikitech.wikimedia.org/wiki/Logstash#Beta_Cluster_Logstash

Be around on IRC

• Add yourself to the "deployer" field of Deployments if you're not already there
• Be online in the libera.chat IRC channel #wikimedia-operations ^connect (and stay online through the deployment window)

Logs to monitor

• Parsoid Cluster in eqiad
• Logstash
• wt2html perf
• html2wt perf

Post-deploy checks

• Test VE editing on enwiki and non-latin wikis
  • For example, open it:Luna (or other complex page), start the visual editor, make some random vandalism, click save -> review changes, then verify that the wikitext reflects your changes and was not corrupted. Hit cancel to abort the edit.
  • Reading through the recent edits (frwiki, enwiki) can also be a good check.

Testing a version bump

If the deployed version of Parsoid updates the Parsoid DOM version and/or will exercises the html2html "down convert" endpoint, the following test procedure will ensure that clients are getting the appropriate DOM version:

• First and foremost, mocha tests should already be present that cover both downgrading the HTML and serializing it with and without selser.
• Create a test page on the beta cluster containing the features that merited the major version bump.
• Deploy the desired commit to the beta cluster and, as a sanity check, make requests for the above test page from Parsoid directly (via deployment-parsoid11.deployment-prep.eqiad.wmflabs) accepting the various specs that are available. The inline meta tag and aforementioned features should indicate that it worked. Example requests might be,
  • For the old version, curl -x deployment-parsoid11:80 'http://en.wikipedia.beta.wmflabs.org/w/rest.php/en.wikipedia.beta.wmflabs.org/v3/page/html/Test_Page' -H'Accept: text/html; charset=utf-8; profile="https://www.mediawiki.org/wiki/Specs/HTML/1.7.0"'
  • For the new version, curl -x deployment-parsoid11:80 'http://en.wikipedia.beta.wmflabs.org/w/rest.php/en.wikipedia.beta.wmflabs.org/v3/page/html/Test_Page' -H'Accept: text/html; charset=utf-8; profile="https://www.mediawiki.org/wiki/Specs/HTML/2.0.0"'
• Confirm that VE on the beta cluster is still tied to the older content version and will be needing a downgrade (see the commit in Special:Version for the extension and compare with the header defined in includes/ApiVisualEditor.php)
• At this point, two scenarios need to be tested: an edit starting from the older content version stored in RESTBase (which won't require a downgrade) and one starting from the new content version, which will.
  • Note that, for extra points, there are potentially several versions numbers stored in RESTBase that satisfy the VE request based on caret semantics and it might be worthwhile to confirm that edits starting from those versions work as well.
  • Once you've found stored content in RESTBase with an appropriate version for your test it's prudent to confirm that VE is actually editing what you expect. This can be achieved by dumping the various DOMs: the original copy(ve.init.target.doc.body.outerHTML) and the edited copy(ve.init.target.docToSave.body.outerHTML)
• In each case, try to confirm that the features can be edited directly as well as being ignored by selser (usually because no normalizations occur). Unfortunately, testing here is a bit more art than science.
• Finally, open up the various testing dashboards for logging and metrics to verify that no unexpected errors are present and that the downgrades are accounted for.

Testing on scandium

When on scandium, use this command to test Parsoid directly:

curl -x scandium.eqiad.wmnet:80 http://<domain>/w/rest.php/<domain>/v3/page/html/<title>/<revid>

Testing LanguageConverter

LanguageConverter can be tested on beta in a manner similar to testing a version bump.

• Create a test page on the beta cluster containing the language converter features you wish to touch. Either the page language for the article must be set to a language w/ variants, or else the article must take place on a wiki where the main language has variants. We'll use the SrTest page on beta srwiki in our examples below.
• Deploy the desired commit to the beta cluster and, as a sanity check, make requests for the above test page from Parsoid directly (via ssh to deployment-parsoid11.deployment-prep.eqiad.wmflabs) specifying the desired variant language. Verify that the result has been converted appropriately. Example requests might be,

• • curl -H'Accept-Language: sr-ec' -x deployment-parsoid11:80 http://sr.wikipedia.beta.wmflabs.org/w/rest.php/sr.wikipedia.beta.wmflabs.org/v3/page/html/User:Cscott%2FSrTest/23
  • curl -H'Accept-Language: sr-el' -x deployment-parsoid11:80 http://sr.wikipedia.beta.wmflabs.org/w/rest.php/sr.wikipedia.beta.wmflabs.org/v3/page/html/User:Cscott%2FSrTest/23
• To test in production, try something like:
  • curl -X GET --header 'Accept-Language: sr-el' 'https://sr.wikipedia.org/api/rest_v1/page/html/%D0%93%D0%BB%D0%B0%D0%B2%D0%BD%D0%B0_%D1%81%D1%82%D1%80%D0%B0%D0%BD%D0%B0/21280369'

See https://phabricator.wikimedia.org/T241146#5810424 for some more examples.

Deploying a cherry-picked patch

One way to do this is to create a new branch in the Parsoid repo and cherry-pick your patches to that. For example:

git checkout v0.13.0-a3 # this is the commit on the master branch that you want to cherry pick on top of
git checkout -b deploy-20150528 # give it a name (go ahead and use the date of your deploy)
git cherry-pick f274c3f54f385a6ac159a47209d279b9040a161c # patch number 1
git cherry-pick de087b106be48fc6e97f2ebc4644f9d297ecdfed # patch number 2
git push gerrit deploy-20150528:deploy-20150528 # create the branch in gerrit (DON'T USE SLASHES HERE)

Now do the usual steps to tag a release and prepare a vendor branch patch (see above) using the next available release version number (v0.13.0-a4 in the example below):

git tag v0.13.0-a4 # this is the next available release number
git push origin v0.13.0-a4

Switch to the mediawiki/vendor repository:

git checkout master ; git pull origin master
edit composer.json # set wikimedia/parsoid to v0.13.0-a4
composer update --no-dev
git add -u
git commit -m "Bump wikimedia/parsoid to v0.13.0-a4"
git review -u

Note that the automated push to beta will fail if your gerrit branch name contains a slash. This is probably just because some ancient version of git is being used, and will eventually be fixed. But in the meantime, use dashes instead of slashes.

When this is merged into mediawiki-vendor it will (shortly) go live on beta; you should verify that everything looks good there. See #Verify deployment version on beta after the vendor patch is merged. If you want this cherry-pick to shortcut the train (instead of waiting to ride the next one) keep going into the next section, "If the train branch has already been cut".

Edge case deployment scenarios

If the train branch has already been cut

IF THE TRAIN BRANCH HAS ALREADY BEEN CUT (aka the wmf/1.XX.0-wmf.YY branch exists) then after you merge to master of mediawiki-vendor you will also need to cherry-pick a patch to the appropriate branch of mediawiki-vendor, for example wmf/1.36.0-wmf.21. In some cases you can use gerrit to cherry-pick the vendor branch to the branch, but in practice most updates to vendor conflict with each other due to the presence of content hashes, so you'll most likely need to repeat the steps above:

# from mediawiki/vendor
git remote update # if needed
git checkout wmf/1.36.0-wmf.3
edit composer.json # set wikimedia/parsoid to v0.13.0-a21
composer update --no-dev
git add -u
git commit -m "Bump wikimedia/parsoid to v0.13.0-a21"
git review -u

Now, before you merge this cherry-pick onto the branch, you need to check one of three possible cases:

1. If the train branch is new and the "branch commit" has not yet been merged (it looks like this; here is a gerrit search) -- wait! Do not merge the cherry-pick into mediawiki-vendor until the branch commit has landed, or the git submodules in mediawiki-core will be left out of sync (T259832). You might want to add a Depends-On clause to the cherry-pick patch to enforce this. If you accidentally merged this, see below for how to fix it.
2. If the branch commit has been merged, but the train has not been deployed anywhere (check Deployments and the status page on versions.toolforge.org), then it's safe to just C+2 the cherry-pick. But be sure to ping #wikimedia-operations ^connect and get clearance before C+2 and merge, since (a) the deployer may have already checked out the branch in preparation for the train, and (b) since jenkins can take a while to complete the merge and they need to know to wait for it. Probably worth leaving a comment on the phab task for the blocker bug for the train release as well.
3. If the train has already been deployed, then you will need to backport this cherry-pick; it is considered bad form to leave code committed on the branch which isn't deployed. Don't merge the cherry-pick until the backport window.

If you accidentally merged into vendor before the branch commit has been merged

Merging a patch onto a branch in the mediawiki-vendor repository will automatically update the git submodules in core, but only after the branch commit is in place. See phab:T259832 for details. If you think you might have merged onto vendor before the branch commit was merged, check the appropriate vendor branch history for core, aka https://gerrit.wikimedia.org/g/mediawiki/core/+/refs/heads/wmf/1.36.0-wmf.3. Verify that the submodule hash for vendor corresponds to the tip of the branch of mediawiki-vendor. If it's not correct, after the branch commit has been merged into mediawiki-core you need to manually bump the submodules:

cd .../mediawiki-core
# note that the below will clobber your vendor, extensions, and skins directories
# you might want to use a new clean checkout of core
git checkout wmf/1.36.0-wmf.3
git submodule update --init
git submodule update --remote vendor
git add vendor
git commit -m "Update git submodules"
git review -u

Review and merge that.

Misc stuff

• To deploy to a single host

  scap deploy --force -l <node>

• To see which hosts are pooled, from another host

  confctl select dc=.*,cluster=parsoid,service=parsoid get

• To see the list of parsoid hosts in beta:

  cat /srv/deployment/parsoid/deploy/scap/betacluster

  • See also /srv/deployment/parsoid/deploy/scap/scap.cfg in general
• To pool/depool a node, from deployment.eqiad.wmnet, run:
  • To depool: SSH_AUTH_SOCK=/run/keyholder/proxy.sock ssh -l deploy-service <node> 'depool service=parsoid'
  • To pool  : SSH_AUTH_SOCK=/run/keyholder/proxy.sock ssh -l deploy-service <node> 'pool service=parsoid'

Data flow

Parsoid runs entirely on an internal subnet, so requests to it are proxied through the ve-parsoid API module. This module is implemented in extensions/VisualEditor/ApiVisualEditor.php and is invoked with a POST request to /w/api.php?action=ve-parsoid. The API module then sends a request to Parsoid, either GET /$prefix/$pagename to get the HTML for a page, or POST /$prefix/$pagename to submit HTML and get wikitext back. Parsoid itself also issues requests to /w/api.php to get the wikitext of the requested page and to do template expansion.

Once the ve-parsoid API module receives a response from Parsoid, it either relays it back to the client (when requesting HTML), or saves the returned wikitext to the page (when submitting HTML).

                (POST /w/api.php?action=ve-parsoid)          (GET /en/Barack_Obama?oldid=1234)           (requests for page content and template expansions)
Client browser ------------------------------------------> API ---------------------------->  Parsoid -----------------------------------------------------> API
    ^                                                      | ^                                 |   ^                                                          |
    |                  (response)                          | |      (HTML)                     |   |                   (responses)                            |
    +------------------------------------------------------+ +---------------------------------+   +----------------------------------------------------------+


                (POST /w/api.php?action=ve-parsoid)          (POST /en/Barack_Obama; oldid=1234)
Client browser ------------------------------------------> API ---------------------------->  Parsoid
                                                           | ^                                 |
                                               (save page) | |      (wikitext)                 |
                                                           | +---------------------------------+
                                                           |
                                                        Database