Grafana Alloy integration #126
Conversation
balanza
force-pushed
the
rfc-alloy-integration
branch
from
ac01daf to
6302009
Compare
balanza
force-pushed
the
rfc-alloy-integration
branch
from
6302009 to
5f4224e
Compare
nelsonkopliku
left a comment
nelsonkopliku
left a comment
There was a problem hiding this comment.
@balanza thanks a lot for this! Great job!
I left some comments, the most concerning one is about reachability and authentication of the prometheus write url
| %endif | ||
| ---- | ||
|
|
||
| NOTE: The `sle_version` macro is provided by OBS and contains the SLES version as an integer (e.g., `150000` for SLES 15, `160000` for SLES 16). The `configure-alloy.sh` script is only installed on SLES 16 and later. |
There was a problem hiding this comment.
praise: great finding! Would you mind sharing the reference to where this is documented please?
| Wants=alloy.service | ||
| After=alloy.service |
There was a problem hiding this comment.
thought:
currently agent spec defines
Recommends: golang-github-prometheus-node_exporterhttps://github.com/trento-project/agent/blob/main/packaging/suse/trento-agent.spec#L33
Would it make sense to have a conditional reccomendation based on SLES version?
if sles < 16
recommends exporter
else if sles >= 16
recommends alloy
| chmod 640 "${ALLOY_CONFIG_FILE}" | ||
|
|
||
| # Reload Alloy to pick up the new configuration | ||
| systemctl reload alloy |
There was a problem hiding this comment.
issue: if the configure-alloy.sh script is going to be executed in ExecStartPre we need to be mindful about the fact that alloy service might not be available and systemctl reload alloy would fail, right?
| facts-service-url: amqp://guest:guest@localhost:5672 | ||
|
|
||
| # Alloy configuration options | ||
| prometheus-url: https://trento.example.com/api/v1/write |
There was a problem hiding this comment.
question:
If I get it right /api/v1/write is what a <prometheus-instance> exposes.
How should the prometheus write api be reachable by alloy?
- directly via the
<prometheus-instance>address? - via trento proxy (nginx/ingress)
Asking because it might have implications for the authentication and I am wondering to what extent trento should be responsible for that authentication.
This RFC should serve as a blueprint for integrating system metrics with Prometheus on Grafana Alloy-based system.
The integration is needed because SLES >= 16 does not ship
node_expoteranymore; instead, the default collector tool is Grafana Alloy.The main goal of the RFC is to reach feature parity with the current solution both in terms of collected metrics and user experience.