bpm is explicit about the interface that it provides to your job. If there is anything in this specification which is unclear or non-specific then please let us know so that we can be explicit about the interface and guarantees provided.
Your process is started and has an unlimited amount of time to start up. You should use a post-start script and a health check if you want your job to only say it has completed deploying after it has started up. You do not need to manage any PID files yourself.
On shutdown your job will receive a
SIGTERM. You then have 20 seconds to
shutdown your application before it will be sent
SIGQUIT to dump the stack
(this is default behavior in the Go and Java runtimes) before being forcibly
If you require longer than this then you should use a drain script for your server. The drain script should put your server in such a state that it can shutdown within 20 seconds. It is acceptable and supported to terminate your process while running the drain script.
Your process should write logs to standard output and standard error file
descriptors. This data will be written to
Any other files which are written to
/var/vcap/sys/log/[job-name] inside the
container will be written to
/var/vcap/sys/log/[job-name] in the host system.
Applications may store temporary files in either
/var/vcap/data/[job-name]/tmp (as per the BOSH recommendation). Both paths,
/var/tmp are mapped onto the same host volume
/var/vcap/data/[job-name]/tmp) and changes made to one will instantly appear
in the other.
bpm will set the
TMPDIR environment variable when execuring your job to one
of the paths listed above. This environment variable is respected by the
majority of in-use standard libraries used by Cloud Foundry.
Note: Per the BOSH team's guidance the temporary filesystem is not mounted as
You should store ephemeral data in
/var/vcap/data/[job-name]. There are no
guarantees that this data will be present in a different invocation of your
job. The lifecycle of this data is tied with the underlying BOSH stemcell.
This storage area is useful for cached data that can be re-created by other means when necessary.
This storage area is shared between all processes in your job. It is your responsibility to make sure that the data saved by each of your job's processes does not collide.
When the path
/var/vcap/store exists bpm will mount the path
/var/vcap/store/[job-name] into the job filesystem. As with
/var/vcap/data/[job-name] bpm will create the leaf directory if it does not
NOTE: Because persistent data access defaults to the
/var/vcap/storeif the job name changes persistent data will no longer be accessible. When changing the name of a job the BOSH
pre-startscript should idempotently move the persistent data from
Note: This feature has been added to enable support for older jobs which use the file system to communicate with other jobs. It is preferable in nearly all cases to use the network to communicate across job boundaries. Using the network reduces scheduling constraints such that a job can be un-collocated and allows more modern access control (key rotation, etc.) of the information being transferred between your jobs.
If your job configuration lists shared volumes (
volumes: key) then these
paths will be created if they do not exist before being mounted into the job
filesystem. These volumes will be mounted read-write. All collocated jobs which
list a particular volume path will be given a volume they can all share.
Only volume paths inside the
/var/vcap/data, and (when present) the
/var/vcap/store directories are currently permitted. Specifying paths which
are not inside this directory will cause the job to fail to start.
There are various
monit quirks that bpm attempts to hide or smooth over.
If an operator runs
monit stop JOB && monit start JOB then
monit will wait
until the monitored process has completely stopped and the
listed in the configuration has completely finished before attempting to start
the job again. This is unsurprising behavior.
monit restart does not wait until the existing
job has stopped before starting the same job again. As soon as the existing job
has removed its PID file
monit forks and starts the new process. This
behavior is unexpected, subtle, and full of race conditions. A considerable
number of BOSH releases are unknowingly suffering from bugs caused by this.
bpm enforces its own locking around process operations to avoid these race
conditions. It is completely safe (from a correctness perspective, you may
still break your service) to run
monit restart on a job which uses bpm.