swarm-otel: key the log stream by machine, not by a hostname every container shares

`_stream_fields=_HOSTNAME,_SYSTEMD_UNIT` was chosen to give "one stream
per unit per machine", which is what the comment above it claims. It does
not: no container sets `networking.hostName`, so `_HOSTNAME` is the NixOS
default `nixos` in every one of them, and every container's stream for a
given unit name merges into a single series.

Measured from inside an agent container:

  journalctl -n1 -o json -> "_MACHINE_ID":"5d1427ea…", "_HOSTNAME":"nixos"
  cat /etc/machine-id    ->  5d1427ea…
  git grep -i hostname -- nix/  -> 20 hits, all public vhost names,
                                   zero `networking.hostName =`

`_MACHINE_ID` is written by journald per machine and is already on every
entry, so adding it to the stream key makes the partition what its own
comment says it is. That merge is also why the host collector and the
containerised one were one series: both log to
`opentelemetry-collector.service`, and the field meant to separate them
was a constant.

The second comment touched here asserted a reader "can still tell the
origins apart" from `_HOSTNAME`/`_SYSTEMD_UNIT`/`_MACHINE_ID`. True only
of the last one; it now says so.

Not a full fix for attribution: `_MACHINE_ID` is opaque hex and nothing
maps it to a container name. Naming is the other half and is a separate
change.

Refs #4304
This commit is contained in:
atlas 2026-09-12 14:14:11 +02:00
commit bb38e1967c
2 changed files with 25 additions and 7 deletions

View file

@ -1106,10 +1106,11 @@ in
# switch, and its `matches` field is an allowlist too, so
# "everything except user sessions" is not expressible.
#
# Attribution is journald's either way — `_HOSTNAME`,
# `_SYSTEMD_UNIT` and `_MACHINE_ID` are written by journald
# rather than by the logging process, so a reader can still
# tell the origins apart.
# Attribution is journald's either way — these fields are
# written by it rather than by the logging process. ⚠️ But
# `_MACHINE_ID` is the only one that separates containers:
# none sets `networking.hostName`, so `_HOSTNAME` is the
# NixOS default `nixos` everywhere but the host itself.
// lib.optionalAttrs collectLogs {
journald = {
directory = hostJournalDir;
@ -1156,10 +1157,15 @@ in
# fields, so without `_msg_field` every record stores the
# literal `missing _msg field` as its message and text search
# finds nothing, while ingest keeps answering 200.
# `_stream_fields` is one stream per unit per machine instead
# of one for the whole host.
# `_stream_fields` is one stream per unit per machine.
# `_MACHINE_ID` is what makes that true: `_HOSTNAME` reads
# `nixos` in every container, so naming it alone merges every
# container's stream for a unit name into one — which is how
# two collectors sharing `opentelemetry-collector.service`
# became one series.
"otlphttp/victorialogs" = {
logs_endpoint = logsPushUrl + "?_msg_field=MESSAGE&_stream_fields=_HOSTNAME,_SYSTEMD_UNIT";
logs_endpoint =
logsPushUrl + "?_msg_field=MESSAGE&_stream_fields=_HOSTNAME,_MACHINE_ID,_SYSTEMD_UNIT";
}
// lib.optionalAttrs haveCollectorSecret {
auth.authenticator = pushAuthenticator "victorialogs";

View file

@ -1437,6 +1437,18 @@ let
&& lib.hasInfix "metrics.t.local" m
&& lib.hasInfix "logs.t.local" l;
}
{
# `_HOSTNAME` cannot separate containers: none sets
# `networking.hostName`, so it is `nixos` for all of them and every
# container's stream for a unit name merges into one.
name = "the log stream is keyed by machine, not only by a hostname every container shares";
ok =
let
l = (otelSettings otelNoStores).exporters."otlphttp/victorialogs".logs_endpoint;
field = f: lib.hasInfix ("_stream_fields=" + f) l || lib.hasInfix ("," + f) l;
in
field "_MACHINE_ID" && field "_SYSTEMD_UNIT" && !(field "_NOSUCHFIELD");
}
{
# The collector reaches these routes through the gateway now, so each
# store needs an ingest location of its own. Without one the write rides