TMPFS weirdness

[bileslav]

Prerequisites

• Have you checked for an existing issue describing your problem?
• Are you running the latest version?
• Is your ports tree recent?
• Is your FreeBSD Host on a supported release?

Describe the bug

The configurations mentioned below are presented at the end of the issue.

1. devel/electron32 builds just fine with the config B, but it fails due to an exhausted memory with the config A.
2. This is what I get using the config B (look at TMPFS usage):

 ID  TOTAL              ORIGIN   PKGNAME           PHASE PHASE    TMPFS        CPU%  MEM%
[01] 00:17:07 devel/electron32 | electron32-32.2.5 build 00:05:21 39.18 GiB 1541.6% 16.4%
[04] 00:17:07      www/firefox | firefox-133.0,2   build 00:15:00 37.93 GiB   91.8%  6.2%
[07] 00:17:07         lang/ghc | ghc-9.6.6_1       build 00:09:55 38.98 GiB  159.2%  2.4%

For what it's worth, I'm almost sure that I didn't observe such numbers using the config A.

How to reproduce

Build devel/electron32 using the A and B configurations.

Expected behavior

1. A successful build in both cases.
2. Real TMPFS usage.

Environment

Host memory	64 GiB
Host OS version	14.1 amd64
Jail OS version	14.1 amd64
Poudriere version	git-3.4.99.20240811
Ports branch	latest

Additional context

Configuration A

USE_TMPFS=all
TMPFS_LIMIT=8
TMPFS_BLACKLIST="rust llvm* electron* ghc* firefox"
TMPFS_BLACKLIST_TMPDIR=${BASEFS}/data/cache/tmp
# MAX_MEMORY isn't set

Configuration B

USE_TMPFS="wrkdir data localbase"
TMPFS_LIMIT=8
TMPFS_BLACKLIST="rust llvm* electron* ghc* firefox"
TMPFS_BLACKLIST_TMPDIR=${BASEFS}/data/cache/tmp
# MAX_MEMORY isn't set

The difference is only in the value of USE_TMPFS.

[bdrewery]

@bileslav How much space do you have on the filesystem? I am guessing that the port build is filling on /tmp in the jail which lands on your host ZFS/UFS.

---

From just reviewing the code I see some problems.

It looks like there are several places that check ${TMPFS_ALL} -eq 1 which do not consider the individual options as an alternative. I think ${TMPFS_ALL} needs to be only checked in 1 place (to enable the smaller flags).

I think the real problem is that we don't have a USE_TMPFS=builder as implied by USE_TMPFS=all doc:

# all       - Run the entire build in memory, including builder jails.

We only have wrkdir, data, and localbase, but no base/builder. And tmpdir apparently.

[bileslav]

@bdrewery,

How much space do you have on the filesystem? I am guessing that the port build is filling on /tmp in the jail which lands on your host ZFS/UFS.

1T, only half occupied. I use ZFS.

We only have wrkdir, data, and localbase, but no base/builder. And tmpdir apparently.

Yes, I also thought about that.

Just in case, I want to point out that I put electron* to TMPFS_BLACKLIST in both configurations.