Slow query log

CamusDB can record statements that take longer than a threshold and expose them as a SQL result set:

SHOW SLOW QUERIES;
SHOW SLOW QUERIES LIKE '%FROM orders%';

The log is off by default and bounded in memory. Use it as a diagnostic while a node is slow, not as an audit log.

Enable it

Configure the node:

slow_query_log_enabled: true
slow_query_log_threshold_ms: 1000
slow_query_log_max_entries: 200
slow_query_log_max_sql_length: 4096

Setting	Default	Meaning
slow_query_log_enabled	false	Master switch.
slow_query_log_threshold_ms	1000	Record statements at or above this duration. 0 records every statement.
slow_query_log_max_entries	200	Entries kept before the oldest is overwritten.
slow_query_log_max_sql_length	4096	Characters of SQL text stored per entry.

SHOW VARIABLES LIKE 'slow_query%' shows the effective values on the node you are inspecting.

What it records

Each entry includes the facts that usually explain why a statement was slow:

• full_scan: the plan read a whole relation instead of seeking through an index.
• spilled: a sort, grouping, distinct, hash join, or row buffer wrote to disk.
• rows_read and rows_returned: reading far more rows than returned usually means the predicate needs a better index.
• outcome: completed, abandoned, or failed.
• error_code: the CamusDB error code when the statement failed.

Rows come back newest first.

Column	Type	Meaning
seq	INT64	Recording order on this node. It keeps increasing after the ring wraps.
started_at	STRING	Statement start time in UTC.
duration_ms	FLOAT64	Wall-clock duration.
database	STRING	Database used by the statement.
user	STRING	Authenticated user, or NULL when auth is disabled.
kind	STRING	Statement kind, such as select, insert, or create_table.
rows_returned	INT64	Rows returned, or rows affected for a mutation.
rows_read	INT64	Rows fetched from storage before filtering.
full_scan	BOOL	Whether any part of the statement read a whole relation.
spilled	BOOL	Whether a blocking operator spilled to disk.
outcome	STRING	completed, abandoned, or failed.
error_code	STRING	Error code for failed statements.
truncated	BOOL	Whether the stored SQL was shortened.
sql	STRING	Statement text, with password literals redacted, up to slow_query_log_max_sql_length.

Read the result

SHOW SLOW QUERIES LIKE '%orders%';

seq	duration_ms	kind	rows_returned	rows_read	full_scan	spilled	outcome	sql
412	4180.2	select	12	2400000	true	false	completed	SELECT * FROM orders WHERE region = 'emea'
407	2210.7	select	500	500	false	true	completed	SELECT * FROM orders ORDER BY total DESC

The first query read many rows to return a few, so it probably needs an index on region. The second query read only what it returned but spilled, so the issue is memory or an unbounded sort. Add a LIMIT, raise the spill threshold, or inspect the plan with EXPLAIN.

SHOW SLOW QUERIES is never recorded in the slow query log. A dashboard or SQL client that polls the log therefore does not erase the history it is reading.

Password literals in CREATE USER ... IDENTIFIED BY ... and ALTER USER ... IDENTIFIED BY ... REPLACE ... are replaced with '***' before an entry is stored. A password passed as a bound parameter never appears in the statement text in the first place.

Dashboard

The operator dashboard has a Slow queries panel over the same data. It shows the newest entries, highlights full scans and spills, and refreshes independently from the rest of the page.

Permissions and limits

SHOW SLOW QUERIES requires a superuser while authentication is enabled. The rows include literal SQL text from statements run by other users, so table-level grants cannot safely narrow the output.

Important limits:

• Entries live in memory only and disappear on restart.
• The result is node-local. Inspect each node separately in a cluster.
• The ring is bounded. If seq advanced by more than the ring capacity between reads, entries were overwritten.
• It is not an audit log. Statements below the threshold are never recorded.

Related

• EXPLAIN
• Query planning
• Spill to disk
• Engine stats
• Configuration