INSERT and ingestion performance Skill

INSERT and ingestion performance

Answers "why are inserts slow, failing, or creating too many parts" from system.query_log, system.part_log, system.query_views_log, system.kafka_consumers, system.processes and system.text_log. Run altinity-expert-clickhouse-connection first if the connection mode, cluster and time window are not yet established.

Query packs

checks.sql — 14 checks: Kafka consumer health and scheduling capacity, running and recent insert activity, part creation rate per table, insert vs merge balance, slow inserts, materialized-view overhead on inserts, failed inserts, batch size distribution, Kafka engine ingestion, Kafka messages in the log, Buffer table flush patterns and current insert-related settings. 2 of them need system.part_log, 1 needs system.query_views_log, 1 needs system.text_log.

How to run the query packs

Read each pack file from this skill's directory (the skill loader prints the directory path).
Run statements one at a time, never a whole file. Statements end with ; and start with a -- @check <id> <title> header; keep the id with its result.
Honor -- @requires: skip the statement when the named table is missing, when keeper is required and the server has no Keeper/ZooKeeper, or when the version condition is not met. List skipped ids with the reason.
Keep {cluster} as written when a cluster macro exists; otherwise apply the connection skill's rewrite rule. Any other {placeholder} is a template variable: substitute a real value first or skip the statement.
On an error, record the check id and the first line of the error, then continue. Only for UNKNOWN_IDENTIFIER, run DESCRIBE TABLE system.<table> and drop the missing column.
A severity column is the verdict for that row. Copy it; do not re-grade.

Deep-dive statements

When a slow insert has a known query_id, break its duration down by materialized view. Substitute {query_id} with the id taken from check ingestion-07 or ingestion-08.

SELECT
    view_name,
    view_duration_ms,
    read_rows,
    written_rows,
    status,
    exception
FROM system.query_views_log
WHERE query_id = '{query_id}'
ORDER BY view_duration_ms DESC;

Interpretation rules

Average written_rows below 1000 per insert is micro-batching: the client is paying part-creation cost per row. Recommend batching to 10k-1M rows per INSERT, or enabling async_insert when the client cannot batch. Check ingestion-10 returns batch_status, not severity; treat "Seriously under-batched" as Major and "Could improve" as Moderate.
Many NewPart events with few MergeParts events in the same window (ingestion-06) means merges are not keeping up with ingestion, not that inserts are too slow. The insert side is healthy; the merge side is the finding.
When system.query_views_log shows view duration dominating the insert duration (ingestion-08), the fix is the materialized view query, not the insert. Name the slowest views and their share of the insert time.
Buffer tables add flush latency of their own: rows are visible only after a flush, and a large Buffer adds memory and a loss window on restart. Flush patterns in ingestion-13 explain insert-to-visibility delay that query_log alone does not show.
Checks marked @requires table:system.part_log are skipped when part_log is disabled. Say so explicitly instead of concluding that part creation is normal; without part_log the part creation rate and the insert/merge balance are unknown.
Failed inserts (ingestion-09) with TOO_MANY_PARTS or "Too many parts" text are a merge backlog symptom, not an insert bug. MEMORY_LIMIT_EXCEEDED on an insert usually comes from a materialized view or from a very large single batch.
Kafka checks (ingestion-01, -02, -11, -12) only summarize here. Rising poll or commit age, or a non-empty exception array, is the trigger to move to the Kafka skill rather than to drill down in this one.

Report format

Header: connection mode, cluster or "single node", ClickHouse version, time window.
Findings: table with columns check, severity, object, evidence, recommendation; one row per finding, Critical first. Evidence quotes the numbers from the result rows.
OK checks: one line listing the check ids that returned no problem rows.
Skipped and failed checks: id and reason or first error line. Never omit this section.
Next steps: skills to load next and immediate actions.

Next skills

Kafka consumer lag, consumer exceptions, rebalances, or consumers above pool size → load skill altinity-expert-clickhouse-kafka
Part creation outpacing merges, TOO_MANY_PARTS, many parts per partition → load skill altinity-expert-clickhouse-merges
Materialized view query itself is slow or reads too much → load skill altinity-expert-clickhouse-reporting
Inserts failing with MEMORY_LIMIT_EXCEEDED or high peak memory → load skill altinity-expert-clickhouse-memory
Partition key too granular, wide tables, or materialized view design problems → load skill altinity-expert-clickhouse-schema
Insert latency tracking disk or write throughput → load skill altinity-expert-clickhouse-storage
Inserts blocked on replicated tables or insert_quorum waits → load skill altinity-expert-clickhouse-replication

Agent Skills: INSERT and ingestion performance

Install this agent skill to your local