build: resource is restricted by Bazel scheduler

[Shikugawa]

Build is not worked parallelly. We can find [Sched] and CPU is not fully used, so that it might be caused by Bazel scheduler behavior. This error can be found after this commit.

shikugawa@machine:~/dev/envoy$ bazel build //test/integration:protocol_integration_test --jobs=60
INFO: Analyzed target //test/integration:protocol_integration_test (0 packages loaded, 0 targets configured).
INFO: Found 1 target...
INFO: From CcCmakeMakeRule external/envoy/bazel/foreign_cc/yaml/include:

INFO: From CcCmakeMakeRule external/envoy/bazel/foreign_cc/event/include:
[3 / 2,332] 60 actions, 1 running
    CcConfigureMakeRule external/envoy/bazel/foreign_cc/gperftools_build/include; 14s linux-sandbox
    [Sched] CcCmakeMakeRule external/envoy/bazel/foreign_cc/nghttp2/include; 24s
    [Sched] CcCmakeMakeRule external/envoy/bazel/foreign_cc/ares/include; 24s
    [Sched] Compiling external/com_google_protobuf/src/google/protobuf/compiler/csharp/csharp_message.cc [for host]; 24s
    [Sched] Compiling external/com_google_protobuf/src/google/protobuf/compiler/java/java_extension_lite.cc [for host]; 24s
    [Sched] Compiling external/com_google_protobuf/src/google/protobuf/compiler/java/java_generator_factory.cc [for host]; 24s
    [Sched] Compiling external/com_google_protobuf/src/google/protobuf/compiler/code_generator.cc [for host]; 24s
    [Sched] Compiling external/com_google_protobuf/src/google/protobuf/compiler/csharp/csharp_repeated_message_field.cc [for host]; 24s ...

[lizan]

I can't reproduce this. Do you have anything in user.bazelrc or ~/.bazerc?

[Shikugawa]

user.bazelrc is only this.

build --config=clang
build --test_output=all

The commits before two weeks ago are worked fine. So It might not be a problem on my environment.

[lizan]

What if you don't specify --jobs?

[Shikugawa]

It may be caused by --experimental_local_memory_estimate in .bazelrc. It might have a confusional bug. So we prefer to destroy it.

[mattklein123]

It may be caused by --experimental_local_memory_estimate in .bazelrc. It might have a confusional bug. So we prefer to destroy it.

Yes I hit this also recently. On my underpowered laptop it decided to only run one job with this set. I haven't seen it as an issue on my bigger cloud machine.

[moderation]

I've commented on this in Slack and at #1861 (comment). The behavior changed in Bazel 3.2.0 and any machine without much memory will scale back to a single CPU. I've had to override this setting in .bazelrc on my RPI4 with 4 cores and 8G of RAM and my Lenovo X1 Carbon with 12 cores and 16G of RAM