Edit this page | Blame

Each shepherd service needs its own guix profile.

UPDATE: genecup is now running as a container on tux02. This is a model for getting rid of guix profiles. Note that we do need to fixate guix-bioinformatics and guix-past channels *per* service. Otherwise updates are impossible.

In time we may move to hosting these services in a guix system container (a VM).

Currently there are a bunch of shepherd services (on tux02, as user shepherd) who are using the global shepherd guix profile and loading guix-bioinformatics at service start time to start their services. This is bad because:

we aren't using per-service guix-bioinformatics worktrees, so the branch changes between service refreshes
service start times are high because we have to rebuild the service (due to above)
services which are restarted are in an "untested" state until they work

Services which need to be migrated:

covid19-pubseq (uses shared (broken!) guix profile, uses shared config directory)(currently running by hand by Pjotr)
hrdp-project (uses shared (broken!) guix profile) (upstream source is gone)
ratspub (uses own guix profile, uses shared guix-bioinformatics)(deprecated(?))
pluto (uses own guix profile, uses shared config directory)(currently disabled)

Services which need to be monitored:

genenetwork1 (uses own guix profile, uses shared config directory)
genome_browser (uses own guix profile, uses shared config directory)
mcron (uses shared guix profile, probably OK)
virtuoso (uses shared guix profile, probably OK)

Services which already have their own profile:

Note that the profiles essentially just contain the guix binary, acting like a time machine. It is used to fire up a guix shell/environment:

bnw
cronjob-gitea
cronjob-pubmed (genecup)
genecup
gitea
ipfs
power
rn6app
singlecell

Each shepherd service needs its own guix profile.

Services which need to be migrated:

Services which need to be monitored:

Services which already have their own profile:

Tags