Any way to control core file location?

Environments
1

If we build an application and run tests, the application might core-dump is a crashing bug has been introduced. I would like to optionally override the core pattern/location so that I can look for core files after execution and at least generate a call stack dump. The pattern on the base linux image seems to be
|/usr/share/apport/apport %p %s %c %d %P but I get permission denied when I try to change it by writing to /proc/sys/kernel/core_pattern. Is there any way I can get access to core files if/when my application crashes. To complicate things further, I’m actually running my tests in a docker container on this host, but my understanding is that docker inherits the host behavior in this regard since they are running in the same kernel.

2

Naturally, you need to write a system-wide setting with sudo. If you are in docker, you may need to place it to a directory mounted with -v to be able to access it on the host.

If that doesn’t help, please link to your build to give more details on what’s happening.

3

I agree, sudo is required for this setting, but I still get permission denied, e.g.:

https://travis-ci.org/mellery451/rippled/jobs/543202564

The full set of commands are:

mkdir -p -m 0777 ${TRAVIS_BUILD_DIR}/cores
sudo echo "${TRAVIS_BUILD_DIR}/cores/%e.%p" > /proc/sys/kernel/core_pattern
docker run \
    -t --env-file /tmp/co.env \
    -v ${TRAVIS_HOME}:${TRAVIS_HOME} \
    -w ${TRAVIS_BUILD_DIR} \
    --ulimit "core=-1" \
    $DOCKER_IMAGE \
    /bin/bash -c 'bin/ci/ubuntu/build-and-test.sh'

…so the core pattern setting happen in travis host, not the container (but it should be used by the container since it runs in the same kernel, I assume)

4

Redirection happens in the non-elevated bash here. The linked question uses tricks like tee, sysctl and bash -c for a reason.

1 Like
5

derp - you’re exactly right…non elevated redir was my issue.