Skip to content

otelcol: ship logs+traces to prod OTLP via upstream role#168

Open
samcm wants to merge 6 commits into
masterfrom
vector-clickhouse-logs
Open

otelcol: ship logs+traces to prod OTLP via upstream role#168
samcm wants to merge 6 commits into
masterfrom
vector-clickhouse-logs

Conversation

@samcm
Copy link
Copy Markdown
Member

@samcm samcm commented May 12, 2026

Adds ethpandaops.general.otelcol_contrib (now provided by the collection) alongside vector. New devnets generated from this template will ship logs + traces to https://otlp.analytics.production.platform.ethpandaops.io, where vmauth tags them as external tier and routes into external.otel_logs / external.otel_traces in ClickHouse.

Vector stays as the Loki shipper for now — same dual-write pattern we kept when migrating blob-devnets, glamsterdam-devnets, and bal-devnets. Earlier vector→clickhouse_logs sink approach (reverted in this PR) is superseded by the otelcol path.

samcm added 2 commits May 12, 2026 11:27
Replaces the loki sink in the vector template config with a remap
transform + http sink that posts to platform's logs-external
clickhouse pipeline at logs-ingest.analytics.production.platform.ethpandaops.io.
Reuses secret_loki credentials since the same VMAuth backend fronts
both ingresses.

New devnets generated from this template will ship logs straight to
clickhouse. This mirrors the rollout already done in glamsterdam-devnet-3
(all hosts) and bal-devnet-6 (bootnode-1) where it has been verified
end-to-end.
Comment thread ansible/inventories/devnet-0/group_vars/all/all.yaml Outdated
samcm added 3 commits May 15, 2026 20:41
This reverts commit 46a8c4a.
New template devnets now run ethpandaops.general.otelcol_contrib alongside
vector. Logs + traces flow to the production OTLP endpoint (vmauth tags
as external tier). Vector stays as the Loki shipper for now; sink renamed
[sinks.out] -> [sinks.loki] for clarity and `otelcol` added to vector's
exclude list so it doesn't loop on its own logs.
@samcm samcm changed the title vector: ship logs to clickhouse logs-ingest instead of loki otelcol: ship logs+traces to prod OTLP via upstream role May 15, 2026
…case)

The docker container parser fails to extract container.name from otelcol's
own log file, leaving the attribute nil. The original container.name regex
filter therefore lets every otelcol-contrib log through, both the first-line
structured records and the stack-trace continuation rows that share no
container metadata.

Adding a body-match clause (github.com/open-telemetry/opentelemetry-collector-contrib
or otelcol-contrib substring) catches both cases and drops the self-emitted
spam at the source.

Verified on blob-devnets prysm-besu-full-1: cluster-wide otelcol error volume
fell from ~1M / 5min to 0 with this clause in place.
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

3 participants