aarch64_windows builds failing C compiler out of heap space on 4Gb RAM node build-docker-win2022-x64-3-intel ??

[andrew-m-leonard]

Recent aarch64_windows build failing on node build-docker-win2022-x64-3-intel:
https://ci.adoptium.net/job/build-scripts/job/jobs/job/jdk/job/jdk-windows-aarch64-temurin/34/consoleFull

20:32:47  ad_x86.cpp
20:32:47  s\build\windows-aarch64-server-release\buildjdk\hotspot\variant-server\gensrc\adfiles\ad_x86.cpp(15109) : fatal error C1002: compiler is out of heap space in pass 2
20:32:47     ... (rest of output omitted)
20:32:47  * For target buildjdk_hotspot_variant-server_libjvm_objs_ad_x86_clone.obj:
20:32:47  ad_x86_clone.cpp
20:32:47  ad_x86.hpp(39110): fatal error C1060: compiler is out of heap space

[sxa]

Re-opening as this seems to be happening on the aarch64 cross-build but not the x64 one. Re-running on the AMD system at https://ci.adoptium.net/view/Failing%20Temurin%20jobs/job/build-scripts/job/jobs/job/evaluation/job/jobs/job/jdk17u/job/jdk17u-evaluation-windows-aarch64-temurin/130/

[sxa]

Noting that in several jdk21u jobs it's also failing during the git fetch operation (not the initial clone of jdk21u) - this is happening after it has processed the cacerts and is fetching tags before doing the build e.g. https://ci.adoptium.net/job/build-scripts/job/jobs/job/jdk21u/job/jdk21u-windows-x64-temurin/221/

16:13:29  build.sh : 16:13:29 : Initiating build ...
16:13:29  Pulling latest tags and getting the latest update version using git fetch -q --tags --depth=1
16:13:29  NOTE: This can take quite some time!  Please be patient
16:15:00  fatal: packfile .git/objects/pack/pack-e80f9bf241984c741f5f0cb24f3db52dbf978e1c.pack cannot be mapped: Resource temporarily unavailable
16:15:00  fatal: fetch-pack: invalid index-pack output

[sxa]

Noting from the previous comments that Andrew's builds where he had seen this were run with different parameters from the ones I'd been testing with which match the ea builds, which may be why we haven't seen it on the ea ones ...

My build here went through to the build itself ok and had:

    "CLEAN_WORKSPACE": false,
    "CLEAN_WORKSPACE_AFTER": false,
    "CLEAN_WORKSPACE_BUILD_OUTPUT_ONLY_AFTER": true

Andrew had the opposite:

    "CLEAN_WORKSPACE": true,
    "CLEAN_WORKSPACE_AFTER": true,
    "CLEAN_WORKSPACE_BUILD_OUTPUT_ONLY_AFTER": false

The ea ones such as this one or last night's jdk17u EA have:

    "CLEAN_WORKSPACE": false,
    "CLEAN_WORKSPACE_AFTER": false,
    "CLEAN_WORKSPACE_BUILD_OUTPUT_ONLY_AFTER": true

So this may be why we're seeing a difference. In the failing case the RAM usage does go up the limit during that git operation:

[sxa]

I've started up another node with 8GiB at https://ci.adoptium.net/computer/dockerhost%2Dazure%2Dwin2022%2Dx64%2D4%2Dintel/ (Spot priced, so Azure may decommission it at any time) if needed to work around such problems for now if we need to run a build with options that are causing problems.