Probably same cause as these two:
https://bugs.chromium.org/p/chromium/issues/detail?id=818880
https://bugs.chromium.org/p/chromium/issues/detail?id=818879

stdout ends with something like:
+------------------------------------------------------------------------+
| Shard 0  https://chromium-swarm.appspot.com/user/task/3c1392a8e889d810 |
+------------------------------------------------------------------------+
+------------------------------------------------------------------------+
| End of shard 0                                                         |
|  Pending: 36046.7s  EXPIRED (lack of capacity)                         |
+------------------------------------------------------------------------+

 Issue 818879  has been merged into this issue.

 Issue 818880  has been merged into this issue.

cwallez@google.com: is this possibly related to the LUCI migration for GPU you mentioned in the LUCI chat earlier today?

We haven't landed any patch yet so it is unrelated.

maruel, vadimsh: any ideas?

I'm looking at the last purple build: https://ci.chromium.org/buildbot/chromium.perf/Win%207%20Perf/1977

All tests that expired there (e.g. https://chromium-swarm.appspot.com/task?id=3c0eb89ef9f88c10&refresh=10&show_raw=1) explicitly target build186-m1. 

Looking at the list of events happening on build86-m1 (https://chromium-swarm.appspot.com/bot?id=build186-m1&selected=1&show_all_events=true&sort_stats=total%3Adesc) I see that the bot was stuck in a bad state (STATUS_DLL_INIT_FAILED errors when running processes, and eventually dead) approx. from 3/2/2018, 4:25:53 AM to 3/5/2018, 4:02:06 PM so it was effectively offline for almost 12h. This is what probably caused all tasks expire there.

Looks like it is working OK now.