New issue
Advanced search Search tips
Note: Color blocks (like or ) mean that a user may not be available. Tooltip shows the reason.

Issue 637628 link

Starred by 3 users

Issue metadata

Status: Assigned
Owner:
Cc:
Components:
EstimatedDays: ----
NextAction: ----
OS: ----
Pri: 1
Type: Bug



Sign in to add a comment

No sheriffs listed on waterfall on MTV weekend (non-APAC weekend)

Project Member Reported by sashab@chromium.org, Aug 15 2016

Issue description

On https://build.chromium.org/p/chromium/waterfall, all sheriffs are listed as "None (channel is sheriff)" on a Monday APAC time, even though the tree still needs to be sheriffed (and does have active sheriffs) at that time.

For example, I was scheduled as build sheriff on an APAC Monday, but I am not listed as an active sheriff on the waterfall.
 
Cc: benhenry@chromium.org jparent@chromium.org estaab@chromium.org
Labels: Infra-Troopers
Owner: ----
Adding the Trooper label.  Not sure who it makes sense to look into this (aka, who owns rotations and displaying them on waterfall ...)
Does this happen reliably every MTV weekend/APAC Monday, or was this the first time you'd seen it happen?
Components: Infra>Platform>Buildbot
Labels: -Infra-Troopers -Pri-3 Pri-2
Owner: hinoka@chromium.org
Status: Assigned (was: Untriaged)
This may be a buildbot bug. Assigning to Ryan who likely has the most context right now.

Comment 4 by hinoka@chromium.org, Aug 17 2016

Cc: hinoka@chromium.org
Owner: mmoss@chromium.org
Not sure if I have any context, but I'm going to kick it over to mmoss who understands the rotation system better than I do.

Comment 5 by mmoss@chromium.org, Aug 17 2016

Status: WontFix (was: Assigned)
There's a cron job that runs on the buildbot masters which reads the sheriff calendars and inserts the shift data into the buildbot html. This cron probably runs based on Pacific time, so probably doesn't display Monday sheriffs until it is actually Monday PST. There is a general lack of timezone awareness in both the calendar entries and our cron jobs, and defaulting to PST is the typical fallback. Any other default would produce similar "incorrect" behavior for other timezones. It's not great, but fixing it for real is complicated and probably not going to be tackled anytime soon (it's been a known, and low priority, issue for basically as long as sheriff calendars have existed, so many years now).
Does wontfix'ing this mean it will also not be fixed in the buildbot replacement? Can this be tracked somewhere else so it doesn't get lost?

Comment 7 by estaab@chromium.org, Aug 17 2016

Components: Infra>Platform>Milo Infra>Sheriffing
Owner: ----
Status: Available (was: WontFix)
We shouldn't close this since it's still an issue for developers.

The rotations pipeline doesn't have a roadmap at the moment and only recently we've formed a sheriffing team and assigned them soft owners of rotations. This falls somewhere between sheriffing and platform, but since platform will need to implement a replacement before turning down buildbot we can take ownership of this when that happens.

If this is something that needs to be fixed in ~1 quarter I suggest the sheriffing team picks this up to fix in the existing rotations code.

Comment 8 by mgiuca@chromium.org, Nov 23 2016

I just realised that Tokyo is one day behind us in Sydney, so this actually only applies to Sydney (not Tokyo).

I'll give a bit more detail about this. There are essentially two timezone related issues for AEST (Sydney time):

1. The on-duty *day* (as given in the emails) is off by one. If the email tells you Tue and Wed, you will actually be on duty on Wed and Thu (i.e., late evening Tue in PST time is business hours on Wed in AEST times, so the day is correct in PST but not in AEST). This should be solved generally by emailing the time range, not date, and letting us figure out which day that is (preferably it would be in the recipient's local time zone, but if not, we can figure it out as long as it gives a time range and time zone).

2. This means a "Friday" sheriff rotation is actually Saturday in Sydney. We don't do Saturday sheriffing so instead we do our Friday shift on a Monday. This gets to the root of what Sasha reported here, which is that during the Monday shift we are not officially the sheriff, so nobody knows who to contact.

I didn't realise the day discrepancy between TOK and SYD until just now (I'm also assuming it's just those two locations that sheriff in APAC; there may be more). This means we may need some more intelligent scheduling, e.g., always make sure someone from TOK is scheduled on Friday, and add a new slot on Sunday and make sure someone from SYD is scheduled then (which will be Monday for us).

Comment 9 by mgiuca@chromium.org, Nov 23 2016

#8 Sigh... sashab@ informed me that TOK is actually the same day as us in SYD.

I based this on a Google search for "time in tok" which, it turns out, was telling me the time in Tok, AK, USA, a town with 1,200 people. http://i0.kym-cdn.com/photos/images/original/000/909/991/48c.jpg

OK so in that case, both #1 and #2 above are relevant to both TOK and SYD. But ignore the last paragraph: instead, we need to move the Friday sheriff slot to Sunday (PST) which would be our Monday.

Comment 10 by no...@chromium.org, Mar 20 2017

Components: Infra>Platform>Milo>Buildbot

Comment 11 by no...@chromium.org, Mar 20 2017

Components: -Infra>Platform>Milo
Components: -Infra>Platform>Buildbot -Infra>Platform>Milo>Buildbot Infra>Documentation
Owner: jparent@chromium.org
Status: Assigned (was: Available)
ping? jparent@ does it make sense to look at this at the same time as the other rotation cleanup work?
Cc: katthomas@chromium.org
Cc: zhangtiff@chromium.org
Labels: -Pri-2 Pri-1
Owner: seanmccullough@chromium.org
By "The other rotation cleanup work" do you mean the Troopers?  Since Troopers and Sheriffs share ZERO common infrastructure now for running the rotation, no, no reason to tie it together.

But, in terms of actually handling this, thank you for surfacing again.

Sean, Tiffany, can you make sure this gets onto Sheriffing team's Q4 OKRs?
I've added "Make sheriff rotation calendars timezone-aware." as a KR under the "Happier Users" O in our Q4 draft. 
> By "The other rotation cleanup work" do you mean the Troopers?  
> Since Troopers and Sheriffs share ZERO common infrastructure now 
> for running the rotation, no, no reason to tie it together.

Correct, but as far as ownership goes, it seems like if you own one you should probably own the other. I am open to counterarguments :).

Components: -Infra>Sheriffing Infra>Sheriffing>Rotations

Comment 19 by olakar@chromium.org, Jan 21 (3 days ago)

Components: Infra>ProdTech>Rotation
ProdTech in Syd is responsible for rotations.

Moving bugs to the new component

Sign in to add a comment