troubleshoot-cockroachdb

# Troubleshoot Cockroachdb ## When to use this skill - **LSM Compaction Death Spiral**: Writes outpace compaction. L0 sublevel count climbs. Read latency goes exponential. Write stalls follow. The node appears "stuck." - **Raft Liveness Failure**: A node becomes slow (GC pause, disk stall, CPU saturation), can't process Raft heartbeats, loses leadership, the cluster redistributes leases causing cascading unavailability windows. - **Clock Skew Crisis**: NTP misconfiguration causes clock drift. Node self-terminates at >80% of max-offset. If multiple nodes drift (common with shared NTP), quorum is lost. - **Hot Range**: A single range receives disproportionate traffic due to sequential key patterns. One node bottlenecks the entire workload while others idle. - **Intent Accumulation**: Long-running or abandoned transactions leave write intents that block other transactions, cascading latency across the cluster. - **Memory Pressure / GC Thrashing**: Large queries exhaust SQL memory budget. Go GC pauses exceed Raft heartbeat interval (3s), causing liveness loss and oscillating availability. - Any time the user reports a Cockroachdb service behaving outside its expected envelope (elevated errors, latency, saturation, resource exhaustion, or unexpected restarts). - An on-call engineer is paging on a Netdata alert tied to a Cockroachdb instance and wants a structured triage path. ## Key facts - This skill wraps the Netdata operator playbook for Cockroachdb. It does not replace the playbook; it routes a coding agent through MCP queries against the same signals the playbook relies on. - CockroachDB is a distributed, strongly-consistent SQL database built on a replicated key-value store. It layers SQL execution on top of a transactional KV engine that uses Raft consensus for replication and MVCC for concurrency control. To reason about its failures, you must hold several interacting subsystems in your head simultaneously. - Dominant failure archetypes the playbook calls out: LSM Compaction Death Spiral; Raft Liveness Failure; Clock Skew Crisis; Hot Range; Intent Accumulation. - Netdata observes the signals listed in the rule files via its native collectors, plus any OpenTelemetry-shipped metrics that your Cockroachdb instrumentation adds. Both paths end at the same MCP query surface. - Netdata's cockroachdb collector emits 60 context(s) under `cockroachdb.*`. The rule files enumerate which contexts surface which domain; the Verification section below names the load-bearing ones explicitly. ## Step-by-step 1. Confirm the Cockroachdb service is up. Query Netdata via MCP with `list_nodes` and filter by the host running the target. A missing node means the symptom is at the network or orchestrator layer, not inside the service. 2. Pull the last 15 minutes of signals for the target. Use `query_metrics` against the contexts listed in the domain rule files. Run `find_anomalous_metrics` in parallel over the same window; anomalies frame which rule file to read first. 3. Check for **LSM Compaction Death Spiral**. Writes outpace compaction. L0 sublevel count climbs. Read latency goes exponential. Write stalls follow. The node appears "stuck." Inspect the rule file whose signals move first for this mode. 4. Check for **Raft Liveness Failure**. A node becomes slow (GC pause, disk stall, CPU saturation), can't process Raft heartbeats, loses leadership, the cluster redistributes leases causing cascading unavailability windows. Inspect the rule file whose signals move first for this mode. 5. Check for **Clock Skew Crisis**. NTP misconfiguration causes clock drift. Node self-terminates at >80% of max-offset. If multiple nodes drift (common with shared NTP), quorum is lost. Inspect the rule file whose signals move first for this mode. 6. Check for **Hot Range**. A single range receives disproportionate traffic due to sequential key patterns. One node bottlenecks the entire workload while others idle. Inspect the rule file whose signals move first for this mode. 7. Check for **Intent Accumulation**. Long-running or abandoned transactions leave write intents that block other transactions, cascading latency across the cluster. Inspect the rule file whose signals move first for this mode. 8. Correlate with host-level signals (`system.cpu.utilization`, `system.memory.usage`, `system.disk.io_time`). Many service-level failures have a host-resource precursor. 9. Apply the remediation hinted at in the matching rule file or the operator playbook. Re-run the MCP queries from the Verification section to confirm the signals returned to expected ranges. A fix that does not move the signal back is not a fix. ### Handy MCP call templates ```text # Discover metrics from Cockroachdb list_metrics with q="cockroachdb" # Pull a specific context over the last window query_metrics with context="cockroachdb.process_uptime", relative_window=-15m # Rank anomalies for the service or host find_anomalous_metrics with node=<host> and context_pattern="cockroachdb.*" # Correlate a known problem context with others find_correlated_metrics around the incident window # Show current alert state list_raised_alerts scoped to the node ``` ## Common mistakes - Treating Cockroachdb as a generic HTTP or process health check. Cockroachdb has specific failure archetypes (see Key facts) that generic checks miss. - Stopping at the first anomalous metric. Several archetypes produce correlated spikes; use `find_correlated_metrics` to widen the search before concluding a root cause. - Quoting percentile latency without the sample count. Low traffic plus a single slow request moves p99 by seconds. - Reading dashboards for a window shorter than the failure's fingerprint. Slow-brew failures (queue growth, bloat, memory fragmentation) need 30+ minutes of data to see the trend. - Skipping the host-level correlation. A process-level fix for a noisy-neighbour problem does not hold. - Assuming alert thresholds are tuned for your workload. Tune against observed Cockroachdb traffic before escalating an alert configuration issue. ## Verification Run these MCP queries against the Netdata instance that sees the Cockroachdb service. Every context listed below is a real Netdata chart name; the agent does not need to guess. ```text 1. list_metrics filtered by q="cockroachdb" (returns every cockroachdb.* context Netdata sees) 2. query_metrics with contexts=[cockroachdb.process_uptime, cockroachdb.sql_connections, cockroachdb.sql_started_dml_statements, cockroachdb.sql_executed_dml_statements, cockroachdb.node_liveness_heartbeats, cockroachdb.sql_errors] and relative_window=-30m 3. find_anomalous_metrics filtered by node=<host> and context_pattern="cockroachdb.*" ``` Load-bearing contexts for this service: - `cockroachdb.process_uptime`: Uptime (seconds). Dimensions: uptime. - `cockroachdb.sql_connections`: Active SQL Connections (connections). Dimensions: active. - `cockroachdb.sql_started_dml_statements`: SQL Started DML Statements (statements). Dimensions: select, update, delete, insert. - `cockroachdb.sql_executed_dml_statements`: SQL Executed DML Statements (statements). Dimensions: select, update, delete, insert. - `cockroachdb.node_liveness_heartbeats`: Node Liveness Heartbeats (heartbeats). Dimensions: successful, failed. - `cockroachdb.sql_errors`: SQL Statements and Transaction Errors (errors). Dimensions: statement, transaction. A clean result means every context is within its expected band and the `find_anomalous_metrics` list is empty or contains only already-acknowledged items. If the fix was real, re-running the same queries 10 minutes after applying it will show a clean result. If it does not, revert and look deeper. ### When the fix does not hold If signals drift back into the anomalous range within 30 minutes of a remediation, the cause was deeper than the applied change. Typical misdiagnoses for Cockroachdb: - Host-resource pressure masquerading as application bug. - Dependent service (DB, cache, upstream) causing a secondary symptom in the instrumented service. - Configuration change that was never reloaded (some subsystems only pick up config on full restart). Escalate by widening the query window: 2-6 hours instead of 15 minutes. Slow-moving causes are invisible at triage window sizes. ## References - [`rules/overview.md`](./rules/overview.md) - Netdata operator playbook: the authoritative source material this skill summarizes. - `skills/netdata-mcp-integration/` for the transport setup. - `skills/netdata-otel-setup/` if additional application signals are needed beyond what Netdata collects natively.