pleroma/otp: always migrate before starting; quiet the default logger

migrate db now always runs (gated only on the binary existing, not on
whether this run freshly installed the shared release) and runs before
the service starts, not after. installed_version only reflects whether
THIS run upgraded the shared /opt/pleroma binary, not whether THIS
instance's own database is migrated against it; gating migrate on that
let every instance but whichever one's play ran first after an upgrade
skip its own migration and crash-loop on "Unapplied Migrations detected"
(hit this for real on nth_io and haskell_social after a 2.11.0 upgrade).
pleroma_ctl migrate is a no-op when nothing's pending, so changed_when
false keeps normal runs from reporting changed every time.

Cap the default console logger handler (journald) at :warning: it had no
level set, so info-level noise like telemetry attach notices and every
SQL debug query were flooding the journal. ex_syslogger's own level was
already set separately; this only affects the handler nothing else was
touching.

2 files changed, 24 insertions, 5 deletions

diff --git a/pleroma/otp/tasks/instance.yaml b/pleroma/otp/tasks/instance.yaml
index a7bfdd4..fa7ca8a 100644
--- a/pleroma/otp/tasks/instance.yaml
+++ b/pleroma/otp/tasks/instance.yaml
@@ -63,10 +63,16 @@
     mode: "0644"
   when: pleroma_fe_theme is defined
 
-- name: ensure pleroma instance is enabled and started
-  become: yes
-  systemd: name="pleroma@{{pleroma_instance}}.service" enabled="yes" state="started"
-
+# Always, not just on a fresh install: /opt/pleroma's release binary is
+# shared across every pleroma instance on a host, but each instance has its
+# own database. installed_version (install.yaml) only reflects whether THIS
+# run installed a new shared release, not whether THIS instance's db has
+# been migrated against it, so gating migrate on that skipped it for every
+# instance but whichever one's play happened to run first after an upgrade,
+# leaving the others crash-looping on "Unapplied Migrations detected".
+# pleroma_ctl migrate is a no-op when nothing's pending, so there's no
+# reason to skip it. Runs before starting the service so an upgrade doesn't
+# briefly crash-loop before migrating.
 - name: check if pleroma binary is present
   stat:
     path: "/opt/pleroma/bin/pleroma_ctl"
@@ -79,4 +85,9 @@
     chdir: "/opt/pleroma/"
   environment:
     PLEROMA_CONFIG_PATH: "/etc/pleroma/{{pleroma_instance}}.config.exs"
-  when: pleroma_binary.stat.exists and not installed_version.stat.exists
+  when: pleroma_binary.stat.exists
+  changed_when: false
+
+- name: ensure pleroma instance is enabled and started
+  become: yes
+  systemd: name="pleroma@{{pleroma_instance}}.service" enabled="yes" state="started"
diff --git a/pleroma/otp/templates/config.exs.j2 b/pleroma/otp/templates/config.exs.j2
index 6f09ec8..f726b2a 100644
--- a/pleroma/otp/templates/config.exs.j2
+++ b/pleroma/otp/templates/config.exs.j2
@@ -71,3 +71,11 @@ config :logger,
 
 config :logger, :ex_syslogger,
   level: :critical
+
+# The above only sets the ex_syslogger backend's level. Logger's separate
+# default console handler (stdout, captured by journald) has no level set
+# and prints everything: info-level telemetry attach notices, every SQL
+# debug query, etc. This caps it at warnings and errors without losing the
+# crash/error output journalctl is actually useful for.
+config :logger,
+  level: :warning