* Switch healthcheck interval/timeout from 30s/3s to 60s/10s.
30s interval is arguably overkill, and 3s timeout is definitely too short
for lower end machines.
* Use HEALTHCHECK CMD exec form to avoid superfluous `sh` invocations.
* Add `--silent --show-error` flags to curl call to avoid progress meter being
shown in healthcheck logs.
The muslrust images seem to have a workdir of /volume as opposed to / in the
others so doing cargo new like this would create the folder in /volume/app.
No need to use two different base images. Debian buster is pulled later
anyway so we can just use it for the vault stage as well.
My reason for this change is partly to avoid redundancy and partly to
make it easier to build everything yourself. When all the build
environment is based on Debian than you just have to figure out how to
build a Debian Docker base image (ref:
https://github.com/ypid/docker-makefile).
This is done to enable backup functionality in the admin interface while
we're waiting for the libsqlite-sys 0.17 to bubble up in the upstream
dependencies. Then we can start using `VACUUM INTO`
This also extends the check for the sqlite binary to also try `sqlite3`
as this is the name of the binary in baseimage distributions we use.
This changes the healthcheck to use `sh` instead of bash, that is absent
from some image versions. (like alpine)
It also removes `*mariadb*` packages from runtime image of sqlite images
as these shouldn't be required.
