Document per-node metric collection in single_endpoint_mode for ClickHouse Cloud#38103
Document per-node metric collection in single_endpoint_mode for ClickHouse Cloud#38103sangeetashivaji wants to merge 7 commits into
Conversation
…House Cloud Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Preview links (active after the
|
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
There was a problem hiding this comment.
More details
This is a documentation-only PR combining related ClickHouse DBM changes across two files: per-node metric collection explanation, Agent version requirement (7.83.0), and Vale style fixes capitalizing "Boolean" in configuration tables. No regressions or factual errors introduced by the PR diff.
🤖 Datadog Autotest · Commit 660d8be · What is Autotest? · Any feedback? Reach out in #autotest
domalessi
left a comment
There was a problem hiding this comment.
Left some editorial suggestions. Let me know if that looks okay to you!
| `single_endpoint_mode: true` is required for ClickHouse Cloud. It enables `clusterAllReplicas()` queries to collect data across all nodes behind a single endpoint. | ||
|
|
||
| When enabled, the Agent also collects standard system-table metrics (from `system.events`, `system.metrics`, `system.asynchronous_metrics`, and `system.errors`) from every node and tags each series with `clickhouse_node`, which produces accurate per-node values. Each node maintains its own cumulative counters, such as the counter behind `clickhouse.query.failed.count`. Without `single_endpoint_mode`, consecutive collections through the load balancer can read these independent per-node counters from different nodes, producing inaccurate metric values. This per-node collection requires Agent version 7.83.0 or later. |
There was a problem hiding this comment.
This paragraph is a bit dense. Splitting up into two paragraphs and breaking up some of the longer sentences.
| `single_endpoint_mode: true` is required for ClickHouse Cloud. It enables `clusterAllReplicas()` queries to collect data across all nodes behind a single endpoint. | |
| When enabled, the Agent also collects standard system-table metrics (from `system.events`, `system.metrics`, `system.asynchronous_metrics`, and `system.errors`) from every node and tags each series with `clickhouse_node`, which produces accurate per-node values. Each node maintains its own cumulative counters, such as the counter behind `clickhouse.query.failed.count`. Without `single_endpoint_mode`, consecutive collections through the load balancer can read these independent per-node counters from different nodes, producing inaccurate metric values. This per-node collection requires Agent version 7.83.0 or later. | |
| `single_endpoint_mode: true` is required for ClickHouse Cloud. It enables `clusterAllReplicas()` queries to collect data across all nodes behind a single endpoint, and helps ensure the Agent reports accurate per-node values for cumulative counters. | |
| Each ClickHouse node maintains its own cumulative counters, such as the one behind `clickhouse.query.failed.count`. Without `single_endpoint_mode`, consecutive collections through the load balancer can read these counters from different nodes, producing inaccurate values. When `single_endpoint_mode` is enabled, the Agent also collects standard system-table metrics (`system.events`, `system.metrics`, `system.asynchronous_metrics`, `system.errors`) from every node and tags each series with `clickhouse_node` to produce accurate per-node values. This requires Agent version 7.83.0 or later. |
|
@sangeetashivaji Looks like you pushed a change after I started reviewing. Let me know if content is set, and I'll look again. Or you can see my suggestion and just apply that if it captures your intent. |
Co-authored-by: domalessi <111786334+domalessi@users.noreply.github.com>
|
domalessi
left a comment
There was a problem hiding this comment.
Two minor suggestions that I'll apply directly. Otherwise, looks good.
Co-authored-by: domalessi <111786334+domalessi@users.noreply.github.com>
|
/merge |
|
View all feedbacks in Devflow UI.
The merge queue is disabled for this repository/branch. |
What does this PR do? What is the motivation?
Clarifies the behavior of
single_endpoint_modeon the ClickHouse Cloud DBM setup page. When enabled, the Agent collects standard system-table metrics (system.events,system.metrics,system.asynchronous_metrics,system.errors) from every node usingclusterAllReplicas()and tags each series withclickhouse_node, producing accurate per-node values for cumulative counters such as the one behindclickhouse.query.failed.count. Without this setting on a load-balanced multi-node service, consecutive collections can read independent per-node counters from different nodes, producing inaccurate metric values.Documents that per-node collection requires Agent version 7.82.0 or later.
Merge instructions
Merge readiness:
Additional notes
Per-node collection for the aggregated tables (
system.parts,system.replicas,system.dictionaries) is intentionally not covered, as those queries useGROUP BY.🤖 Generated with Claude Code