Ben Wagner a3f79c0b42 Update trooper doc with chrome-infra chat.

BUG=skia:

GOLD_TRYBOT_URL= https://gold.skia.org/search?issue=3745
NOTRY=true
DOCS_PREVIEW= https://skia.org/?cl=3745

Change-Id: Icc9b52a7b4bc90f4ef8d8b952d49663f74104ba9
Reviewed-on: https://skia-review.googlesource.com/3745
Reviewed-by: Ravi Mistry <rmistry@google.com>
Commit-Queue: Ben Wagner <benjaminwagner@google.com>

2016-10-24 16:14:01 +00:00

5.6 KiB

Raw Blame History

Infra Trooper Documentation

What does an Infra trooper do?
View current and upcoming troopers
How to swap trooper shifts
Tips for troopers

What does an Infra trooper do?

The trooper has two main jobs:

Keep an eye on Infra alerts emails (sent to infra-alerts@skia.org). The alerts are also available here.
Resolve the above alerts as they come in.

View current and upcoming troopers

The list of troopers is specified in the skia-tree-status web app. The current trooper is highlighted in green. The banner on the top of the status page also displays the current trooper.

How to swap trooper shifts

If you need to swap shifts with someone (because you are out sick or on vacation), please get approval from the person you want to swap with. Then send an email to skiabot@google.com and cc rmistry@.

Tips for troopers

Make sure you are a member of MDB group chrome-skia-ninja. Valentine passwords and Chrome Golo access are based on membership in this group.
These alerts generally auto-dismiss once the criteria for the alert is no longer met:
- Monitoring alerts, including prober, collectd, and others
- Disconnected build slaves
These alerts generally do not auto-dismiss (issue here):
- Build slaves that failed a step
- Disconnected devices (these are detected as the "wait for device" step failing)
"Failed to execute query" may show a different query than the failing one; dismiss the alert to get a new alert showing the query that is actually failing. (All "failed to execute query" alerts are lumped into a single alert, which is why the failed query which initially triggered the alert may not be failing any more but the alert is still active because another query is failing.)
Where machines are located:
- Machine name like "skia-vm-NNN", "ct-vm-NNN" -> GCE
- Machine name ends with "a3", "a4", "m3" -> Chrome Golo
- Machine name ends with "m5" -> CT bare-metal bots in Chrome Golo
- Machine name starts with "skiabot-" -> Chapel Hill lab
- Machine name starts with "win8" -> Chapel Hill lab (Windows machine names can't be very long, so the "skiabot-shuttle-" prefix is dropped.)
- slave11-c3 is a Chrome infra GCE machine (not to be confused with the Skia Buildbots GCE, which we refer to as simply "GCE")
The chrome-infra hangout is useful for questions regarding bots managed by the Chrome Infra team and to get visibility into upstream failures that cause problems for us.
To log in to a Linux buildbot in GCE, use gcloud compute ssh default@<machine name>. Choose the zone listed for the GCE VM (or specify it using the --zone command-line flag).
To log in to a Windows buildbot in GCE, use Chrome RDP Extension with the IP address of the GCE VM shown on the host info page for that bot. The username is chrome-bot and the password can be found on Valentine as "chrome-bot (Win GCE)".
If there is a problem with a bot in the Chrome Golo or Chrome infra GCE, the best course of action is to file a bug with the Chrome infra team. But if you know what you're doing:
- To access bots in the Chrome Golo, follow these instructions.
  - Machine name ends with "a3" or "a4" -> ssh command looks like ssh build3-a3.chrome
  - Machine name ends with "m3" -> ssh command looks like ssh build5-m3.golo
  - Machine name ends with "m5" -> ssh command looks like ssh build1-m5.golo. Example bug to file to Infra Labs.
  - For MacOS and Windows bots, you will be prompted for a password, which is stored on Valentine as "Chrome Golo, Perf, GPU bots - chrome-bot".
- To access bots in the Chrome infra GCE -> command looks like gcutil --project=google.com:chromecompute ssh --ssh_user=default slave11-c3 (or use the ccompute ssh script from the infra_internal repo).
Read over the SkiaLab documentation for more detail on dealing with device alerts.
To stop a buildslave for a device, log in to the host for that device, cd ~/buildbot/<slave name>/build/slave; make stop. To start it again, TESTING_SLAVENAME=<slave name> make start.
Buildslaves can be slow to come up after reboot, but if the buildslave remains disconnected, you may need to start it manually. On Mac and Linux, check using ps aux | grep python that neither buildbot nor gclient are running, then run ~/skiabot-slave-start-on-boot.sh.
Sometimes iOS builds fail with 'The service is invalid'. Try rebooting the iOS host to fix this.

5.6 KiB Raw Blame History