On Jun 16, 2025, at 04:50, Ronald Klop <[email protected]> wrote:

> Van: Mark Millard <[email protected]>
> Datum:vrijdag, 30 mei 2025 20:58
> Aan:"Dag-Erling Smørgrav" <[email protected]>
> CC:[email protected], FreeBSD Mailing List <[email protected]>
> Onderwerp:Re: error -9 on build cluster?
> On May 30, 2025, at 11:29, Mark Millard <[email protected]> wrote:
> 
> > On May 30, 2025, at 10:43, Dag-Erling Smørgrav <[email protected]> wrote:
> >
> >> Mark Millard <[email protected]> writes:
> >>> Dag-Erling Smørgrav <[email protected]> writes:
> >>>> This is preceded by an 800-line list of warnings from a single source
> >>>> file (multi_index_block.cpp). This list is clearly truncated as it's
> >>>> missing the final “N warnings generated” line, so my guess is there is
> >>>> at least one error there as well which just doesn't get displayed.
> >>> My guess: the failure occurred while the warnings were otherwise
> >>> still being generated and that is what stopped the warnings.
> >>
> >> That's possible.  If we assume that the error code returned by scons is
> >> an errno, it corresponds to EBADF.  Perhaps a bug in scons causes it to
> >> lose track of the pipe it uses to read the compiler's stderr.
> >
> > devel/scans has not been updated after 2024-Jun-06:
> >
> > QUOTE
> > author Po-Chuan Hsieh <[email protected]> 2024-06-06 22:03:56 +0000
> > committer Po-Chuan Hsieh <[email protected]> 2024-06-06 22:03:56 +0000
> > . . .
> > devel/scons: Update to 4.7.0
> > Changes: https://github.com/SCons/scons/releases 
> > https://scons.org/tag/releases.html PR: 279488 Exp-run by: antoine
> > END QUOTE
> >
> > That would mean that quarterly got the update in 2024 as well.
> >
> > https://pkg-status.freebsd.org/ampere1/data/141arm64-quarterly/2e2d801cf977/logs/mongodb60-6.0.18.log
> >
> > did not fail but shows the warnings.
> >
> > devel/scons does not seem to be the source of the change in beavior.
> 
> 
> Worked:
> 
> https://pkg-status.freebsd.org/ampere3/data/134arm64-default/1afdc808e67b/logs/mongodb60-6.0.20_1.log
> build started at Sun Jan  5 17:35:49 UTC 2025
> 
> The next 134arm64-default build Failed:
> 
> https://pkg-status.freebsd.org/ampere3/data/134arm64-default/86e28c84a7e0/logs/errors/mongodb60-6.0.20_1.log
> build started at Sun Jan 19 11:21:16 UTC 2025
> 
> That at least  gives a time frame to investigate.
> 
> More details:
> 
> =>> Building databases/mongodb60
> build started at Sun Jan  5 17:35:49 UTC 2025
> port directory: /usr/ports/databases/mongodb60
> package name: mongodb60-6.0.20_1
> building for: FreeBSD 134arm64-default-job-06 13.4-RELEASE-p2 FreeBSD 
> 13.4-RELEASE-p2 arm64
> maintained by: [email protected]
> Makefile datestamp: -rw-r--r--  1 root  wheel  5845 Jan  4 01:02 
> /usr/ports/databases/mongodb60/Makefile
> Ports top last git commit: 1afdc808e67
> Ports top unclean checkout: no
> Port dir last git commit: 3ab18c284b0
> Port dir unclean checkout: no
> Poudriere version: poudriere-git-3.4.2
> Host OSVERSION: 1500028
> Jail OSVERSION: 1304000
> 
> vs.
> 
> =>> Building databases/mongodb60
> build started at Sun Jan 19 11:21:16 UTC 2025
> port directory: /usr/ports/databases/mongodb60
> package name: mongodb60-6.0.20_1
> building for: FreeBSD 134arm64-default-job-13 13.4-RELEASE-p2 FreeBSD 
> 13.4-RELEASE-p2 arm64
> maintained by: [email protected]
> Makefile datestamp: -rw-r--r--  1 root  wheel  5845 Jan  4 01:02 
> /usr/ports/databases/mongodb60/Makefile
> Ports top last git commit: 86e28c84a7e
> Ports top unclean checkout: no
> Port dir last git commit: 3ab18c284b0
> Port dir unclean checkout: no
> Poudriere version: poudriere-git-3.4.2
> Host OSVERSION: 1500028
> Jail OSVERSION: 1304000
> 
> Same Host and Jail OSVERSION's. Same "Port dir last git commit".
> 
> But different "Ports top last git commit". So. some change in
> another package that is involved?
> 
> ===
> Mark Millard
> marklmi at yahoo.com
>   
> 
> 
> Hi,
> 
> The builder got the same error on mongodb70.

Well recent 60 vs. the 70:

scons: *** [build/59f4f0dd/mongo/db/catalog/multi_index_block.o] Error -9
vs.:
scons: *** [build/opt/mongo/db/query/canonical_query_encoder.o] Error -9

It is not at an analogous point in the build. Looking at the build times:

build time: 00:45:05
vs.:
build time: 05:05:52


Looking at the 60 history . . .

scons: *** [build/59f4f0dd/mongo/db/catalog/multi_index_block.o] Error -9
scons: building terminated because of errors.
build/59f4f0dd/mongo/db/catalog/multi_index_block.o failed: Error -9

scons: *** [build/59f4f0dd/mongo/db/views/view_catalog_helpers.o] Error -9
. . .
scons: building terminated because of errors.
build/59f4f0dd/mongo/db/views/view_catalog_helpers.o failed: Error -9

(The above 2 are amd64, the below older ones are all arm64:)

src/mongo/db/commands/collection_to_capped.cpp
. . .
*** Signal 9

build/59f4f0dd/mongo/db/commands/feature_compatibility_version_document_gen.cpp
. . .
*** Signal 9

build/59f4f0dd/mongo/db/storage/storage_parameters_gen.cpp
. . .
*** Signal 9

(The older ones only have the report available now, not the log:)

scons: building terminated because of errors.
build/59f4f0dd/mongo/db/s/config/sharding_catalog_manager_zone_operations.o 
failed: Error -9

scons: building terminated because of errors.
build/59f4f0dd/mongo/db/commands/mr_common.o failed: Error -9

scons: building terminated because of errors.
build/59f4f0dd/mongo/db/query/find.o failed: Error -9

src/third_party/mozjs/platform/aarch64/freebsd/build/Unified_cpp_js_src16.cpp
*** Signal 9

(The above goes back to 2025-02-05 17:50.)

The first is 6.0.23_1. The 2nd is 6.0.22_1. The last 2 are 6.0.20_1.
The rest are 6.0.20_2.

There is no uniformity to where the errors occur. There is also
variability for "Signal 9" vs. "Error -9". A guess would be that
"Error -9" is another way of reporting "Signal 9".

Back in 2022-Jul there was this to avoid "Signal 9" problems:

A commit in branch main references this bug:

URL: 
https://cgit.FreeBSD.org/ports/commit/?id=0aabc02be18794a56c436233e59379ebaa86c2e0

commit 0aabc02be18794a56c436233e59379ebaa86c2e0
Author: Daniel Engberb <[email protected]>
AuthorDate: 2022-07-14 14:43:02 +0000
Commit: Fernando Apesteguía <[email protected]>
CommitDate: 2022-07-16 15:11:58 +0000

databases/mongodb50: Disable LTO by default.

Building mongo with LTO enabled takes a ridiculously amount of memory
(32GB in amd64).
Disable -ggdb so memory requirements don't sky rocket (around 10-12GB)

PR: 264340
Reported by: [email protected]
Approved by: [email protected] (maintainer)

databases/mongodb50/files/patch-SConstruct | 15 +++++++++++----
1 file changed, 11 insertions(+), 4 deletions(-)


There was also a comment reporting console log text like:

swap_pager: out of swap space
swp_pager_getswapspace(2): failed
swap_pager: out of swap space
swp_pager_getswapspace(2): failed
swp_pager_getswapspace(5): failed
swp_pager_getswapspace(1): failed
swp_pager_getswapspace(23): failed
pid 73953 (ld.lld), jid 0, uid 0, was killed: failed to reclaim memory

But, to my knowledge, such information is not publicly
available and would be not easily tied to match the
timing of mongodb* build failures.

So personal/local builds may be needed to figure out if
such might be involved.

> But this build does not have all the warnings. I still don't see the cause of 
> Error -9.
> 
> https://pkg-status.freebsd.org/beefy22/data/142amd64-default/02e9dc3b903b/logs/mongodb70-7.0.21_1.log
> 
> https://portsfallout.com/fallout?port=databases%2Fmongodb70%24
> 
> My next idea is that this occurs because a process gets killed (-9), which 
> should be visible in the logs of the cluster maintainers. Does this makes 
> sense?

===
Mark Millard
marklmi at yahoo.com


Reply via email to