This is what I can see in the builder:

root@gobuilder:~ # du -hs /tmp/*
 88K    /tmp/gobuilder.log
1.3G    /tmp/workdir-host-dragonfly-amd64-master
root@gobuilder:~ # df -h /tmp
Filesystem   Size   Used  Avail Capacity  Mounted on
tmpfs       1967M  1363M   603M    69%    /tmp

<2GiB seems awfully small for a /tmp partition. Any chance it could be increased?

Barring that, any idea what changed around Dec. 7 to cause this builder to start running over the limit? It looks like the failures overwhelmingly occur when linking golang.org/x/website/cmd/golangorg.test or tests closely before or after that one, but there isn't an obvious culprit CL –- it isn't even clear to me whether this is due to a change in x/website proper, or in the size of the linker inputs or outputs due to some other change in the toolchain.

Marking as release-blocker for Go 1.18 via #11811. This is clearly a mid-cycle regression of some sort, and it's not obvious to me what the root cause is or how it can be mitigated.

We did not do anything on that VM, in fact it's up for 70 days and running a kernel from Oct, 17th:

root@gobuilder:~ # uname -a
DragonFly gobuilder.muscles.dragonflybsd.org 6.1-DEVELOPMENT DragonFly v6.1.0.477.gc2e50-DEVELOPMENT #0: Sun Oct 17 10:27:56 UTC 2021     root@pkgbox64.dragonflybsd.org:/usr/obj/usr/src/sys/X86_64_GENERIC  x86_64
root@gobuilder:~ # uptime
 8:25AM  up 70 days,  9:39, 1 user, load averages: 5.79, 4.98, 4.47

The /tmp size can be increased indeed. I was also thinking in upgrading to latest bleeding-edge but I'll make no changes until you tell me to.

Since the dragonfly builder is reverse, @dmitshur suggested that you could go ahead and increase the /tmp partition on this builder and we can investigate the size increases independently (for example, on a GCE builder).

I would suggest not upgrading the OS image at this point in the cycle, though. (Early in the 1.19 development window would give more time to chase down any regressions that may occur.)

@bcmills ok, i'll stop the builder, increase the /tmp partition and restart the VM. I won't be upgrading until you guys tell me to do so. With regards to a GCE builder, I still have to create images for GCE, unless somebody beats me to it.

@tuxillo Thanks! It looks like the builder has been missing for a few days. How is this going?

Sorry! I forgot to start the builder again.

It has now 4GB in /tmp:

root@gobuilder:~ # df -h /tmp
Filesystem   Size   Used  Avail Capacity  Mounted on
tmpfs       4096M   758M  3338M    19%    /tmp

Looks like these failures are no longer occurring. Thanks!