Release Infrastructure SOP

From FedoraProject

Revision as of 00:18, 26 March 2009 by Mdomsch (Talk | contribs)

Jump to: navigation, search
Infrastructure InfrastructureTeamN1.png
Shortcut:
ISOP:RELEASE

This SOP contains all of the steps required by the Fedora Infrastructure team in order to get a release out. Much of this work overlaps with the Release Engineering team (and at present share many of the same members). Some work may get done by releng, some may get done by Infrastructure, as long as it gets done, it doesn't matter.

Contents

Contact Information

Owner: Fedora Infrastructure Team, Fedora Release Engineering Team

Contact: #fedora-admin, #fedora-devel, sysadmin-main, sysadmin-releng group

Location: N/A

Servers: All

Purpose: Releasing a new version of Fedora

Description

Preparations

After test1 for a release ships, the following items need to be completed. *Please open a ticket for all of them individually*

  1. New website from the websites team (typically hosted at http://fedoraproject.org/_/), edit syncStatic.sh to freeze the website in preparation for release updates.
  2. Verify mirror space (for all test releases as well)
  3. Release day ticket and milestone (keep a log of who is doing what and notes)
  4. Verify with rel-eng permissions on content are right on the mirrors. Don't leak.
  5. Add MirrorManager repository redirects from to-be-released version to rawhide repos. This directs pre-release users at rawhide. (all releases, Alpha, Beta, Preview, and Final)
  6. Communication with Red Hat IS (Give at least 2 months notice, then reminders as the time comes near) (final release only)
  7. Infrastructure change freeze
  8. Modify wiki frontpage Common issues redirect to new release content. (final release only)
  9. Re-review Lessons Learned pages from previous release, be sure we aren't forgetting things we meant to do after the last release.
  10. Add a new version to the release in MirrorManager - http://fedoraproject.org/wiki/Infrastructure/SOP/MirrorManager#Release_Preparation


Change Freeze

The rules are simple. If we are in a pre release freeze or a full freeze, take a look at the architecture document. If the hostname is in the freeze area, that entire host is frozen. So making global changes is also forbidden. If, for some reason, a change needs to be made, a request must go to the fedora-infrastructure-list requesting it. It will require at least two people to signoff on the change from either sysadmin-main or the releng group. Exceptions to this include:

  • Planned changes as part of the release
  • Emergencies / outages

Change freezes will be sent to the fedora-infrastructure-list and begin 2 weeks before each release and the final release.

To see the latest environments map run:

sudo yum -y install openoffice.org-draw
git clone git://git.fedorahosted.org/git/fedora-infrastructure.git/
cd fedora-infrastructure/architecture
oodraw Environments.odg

Static / cached urls

The following URL's should be setup to be cached via a static page or with mod_cache_mem as needed. These are the links that will be hit hardest during the release.

  1. http://get.fedoraproject.org/
  2. http://help.fedoraproject.org/
  3. http://join.fedoraproject.org/
  4. http://fedoraproject.org/wiki/Releases/21/ReleaseOverview
  5. http://fedoraproject.org/wiki/Bugs/F21Common
  6. http://docs.fedoraproject.org/release-notes/index.html
  7. http://docs.fedoraproject.org/release-notes/f21/index.html
  8. http://docs.fedoraproject.org/release-notes/f21/en_US/index.html
  9. http://docs.fedoraproject.org/install-guide/index.html
  10. http://docs.fedoraproject.org/install-guide/f21/index.html
  11. http://docs.fedoraproject.org/install-guide/f21/en_US/index.html
  12. http://fedoraproject.org/wiki/Releases/21/ReleaseSummary
  13. http://docs.fedoraproject.org/

Each one of these tasks except for the release day ticket must be completed and closed prior to the release.

Notes about release day

Release day is always an interesting and unique event. After the final sprint from test to the final release a lot of the developers will be looking forward to a bit of time away, as well as some sleep. Once Release Engineering has built the final tree, and synced it to the mirrors it is our job to make sure everything else (except the bit flip) gets done as painlessly and easily as possible.

All communication is typically done in #fedora-admin though we may start also having communications in sip:infrastructure@fedoraproject.org. Typically these channels are laid back and staying on topic isn't strictly enforced. On release day this is not true. We encourage people to come, stay in the room and be quiet unless they have a specific task or question releated to release day. Its nothing personal, but release day can get out of hand quick.

During normal load, our websites function as normal. This is especially true since we've moved the wiki to mod_fcgi. On release day our load spikes a great deal. During the Fedora 6 launch many services were offline for hours. Some (like the docs) were off for days. A large part of this outage was due to the wiki not being able to handle the load, part was a lack of planning by the Infrastructure team, and part is still a mystery. (There are questions as to whether or not all of the traffic was legit or a ddos.

The Fedora 7 release went much better. Some services were offline for minutes at a time but very little of it was out longer then that. The wiki crashed, as it always does. We had made sure to make the fedoraproject.org landing page static though. This helped a great deal though we did see load on the proxy boxes as spiky.

Day Prior to Release Day

Step 1 (Torrent)

Setup the torrent. All files can be synced with the torrent box (exists at duke) but just not published to the world. Verify with sha1sum. Follow the instructions on the http://fedoraproject.org/wiki/Infrastructure/SOP/TorrentRelease page up to and including step 4.

Step 2 (Website)

Verify the website design / content has been finalized with the websites team.

Step 3 (Mirrors)

Verify enough mirrors are setup and have Fedora ready for release. If for some reason something is broken it needs to be fixed. Many of the mirrors are running a check-in script. This lets us know who has Fedora without having to scan everyone.

Step 4 (Enable Updates)

The new Fedora release will likely have 0day updates. Making sure we can push those updates is important.

Remove MirrorManager repository redirects from to-be-released version to rawhide repos. This lets people get the proper updates.

As far as bodhi is concerned

  • Create the new release in bodhi, to allow developers to queue up 0day updates. Make sure this is 'locked' so they don't get pushed. See the bodhi administration page for instructions.
  • Create mash config files for the new release
  • Make sure the dist-fX-updates{,-testing} tags are creating in Koji
  • Make sure the rsync script is updated
  • When it's time to push, simply unlock the release in bodhi and you should be good to go.

Step 5 (Lessons Learned page)

Set up a new page like to record findings we'll use next release cycle.

Release day

Step 1 (Prep and wait)

Verify the mirrors are ready and that the torrent has valid copies of its files (use sha1sum)

Template:Message/Warning

Step 2 (Torrent)

Once given the ok to release, the Infrastructure team should publish the torrent and encourage people to seed. Complete the steps on the http://fedoraproject.org/wiki/Infrastructure/SOP/TorrentRelease page after step 4. Also, upload the files to the SixXS IPv6 torrent seeds, using the form linked from http://www.sixxs.net/tools/tracker/

Step 3 (Bit flip)

The mirrors sit and wait for a single permissions bit to be altered so that they show up to their services. The bit flip (done by the releng team) will replicate out to the mirrors. Verify that the mirrors have received the change by seeing if it is actually available, just use a spot check. Once that is complete move on.

Step 4 (Website)

Once all of the distribution pieces are verified (mirrors and torrent), all that is left is to publish the website. At present this is done by making sure the website team (or someone) tags what is currently in HEAD as LIVE. It will sync in an hour normally but on release day people don't like to wait that long so do the following:

/etc/cron.hourly/syncStatic.sh
/usr/local/bin/syncFiles.sh fedoraproject.org /srv/web/fedoraproject.org/

Verify http://fedoraproject.org/ is working.

Step 5 (Docs)

Just as with the website, the docs site needs to be published. Just as above follow the following steps:

/root/bin/docs-sync

Step 6 (Monitor)

Once the website is live, keep an eye on various news sites for the release announcement. Closely watch the load on all of the boxes, proxy, application and otherwise. If something is getting overloaded, see suggestions on this page in the "Juggling Resources" section.

Step 7 (Done)

Just chill, keep an eye on everything and make changes as needed. If you can't keep a service up, try to redirect randomly to some of the mirrors.

Priorities

Priorities of during release day (In order):

  1. Website - Anything related to a user landing at fedoraproject.org, and clicking through to a mirror or torrent to download something must be kept up. This is distribution, and without it we can potentially lose many users.
  2. Linked addresses - We do not have direct control over what Digg, Slashdot or anyone else links to. If they link to something on the wiki and it is going down or link to any other site we control a rewrite should be put in place to direct them to http://fedoraproject.org/get-fedora.
  3. Torrent - The torrent server has never had problems during a release. Make sure it is up.
  4. Release Notes - Typically grouped with the docs site, the release notes are often linked to (this is fine, no need to redirect) but keep an eye on the logs and ensure that where we've said the release notes are, that they can be found there. In previous releases we sometimes had to make this available in more than one spot.
  5. docs.fedoraproject.org - People will want to see whats new in Fedora and get further documentation about it. Much of this is in the release notes.
  6. wiki - Because it is so resource heavy, and because it is so developer oriented we have no choice but to give the wiki a lower priority.
  7. Everything else.

Juggling Resources

In our environment we're running different things on many different servers. Using Xen we can easily give machines more or less ram, processors. We can take down builders and bring up application servers. The trick is to be smart and make sure you understand what is causing the problem. These are some tips to keep in mind:

  • IPTables based bandwidth and connection limiting (successful in the past)
  • Altering the weight on the proxy balancers (see: Balancers )
  • Create static pages out of otherwise dynamic content
  • Redirect pages to a mirror
  • Add a server / remove un-needed servers