Add 5-minute AggregatingMergeTree stats rollup for dashboard queries
All checks were successful
CI / build (push) Successful in 1m14s
CI / docker (push) Successful in 42s
CI / deploy (push) Successful in 30s

Pre-aggregate route execution stats into 5-minute buckets using a
materialized view with -State/-Merge combinators. Rewrite stats() and
timeseries() to query the rollup table instead of scanning the wide
base table. Active count remains a real-time query since RUNNING is
transient. Includes idempotent backfill migration.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
This commit is contained in:
hsiegeln
2026-03-15 10:46:26 +01:00
parent 6f7c92f793
commit adf13f0430
3 changed files with 97 additions and 44 deletions

View File

@@ -96,79 +96,87 @@ public class ClickHouseSearchEngine implements SearchEngine {
@Override
public ExecutionStats stats(Instant from, Instant to, String routeId, List<String> agentIds) {
// Current period — read from rollup
var conditions = new ArrayList<String>();
var params = new ArrayList<Object>();
conditions.add("start_time >= ?");
conditions.add("bucket >= toStartOfFiveMinutes(?)");
params.add(Timestamp.from(from));
conditions.add("start_time <= ?");
conditions.add("bucket <= ?");
params.add(Timestamp.from(to));
addScopeFilters(routeId, agentIds, conditions, params);
String where = " WHERE " + String.join(" AND ", conditions);
String aggregateSql = "SELECT count() AS total_count, " +
"countIf(status = 'FAILED') AS failed_count, " +
"toInt64(ifNotFinite(avg(duration_ms), 0)) AS avg_duration_ms, " +
"toInt64(ifNotFinite(quantile(0.99)(duration_ms), 0)) AS p99_duration_ms, " +
"countIf(status = 'RUNNING') AS active_count " +
"FROM route_executions" + where;
String rollupSql = "SELECT " +
"countMerge(total_count) AS total_count, " +
"countIfMerge(failed_count) AS failed_count, " +
"toInt64(ifNotFinite(sumMerge(duration_sum) / countMerge(total_count), 0)) AS avg_duration_ms, " +
"toInt64(ifNotFinite(quantileTDigestMerge(0.99)(p99_duration), 0)) AS p99_duration_ms " +
"FROM route_execution_stats_5m" + where;
// Current period
record PeriodStats(long totalCount, long failedCount, long avgDurationMs, long p99LatencyMs, long activeCount) {}
PeriodStats current = jdbcTemplate.queryForObject(aggregateSql,
record PeriodStats(long totalCount, long failedCount, long avgDurationMs, long p99LatencyMs) {}
PeriodStats current = jdbcTemplate.queryForObject(rollupSql,
(rs, rowNum) -> new PeriodStats(
rs.getLong("total_count"),
rs.getLong("failed_count"),
rs.getLong("avg_duration_ms"),
rs.getLong("p99_duration_ms"),
rs.getLong("active_count")),
rs.getLong("p99_duration_ms")),
params.toArray());
// Previous period (same window shifted back 24h)
// Active count — lightweight real-time query on base table (RUNNING is transient)
var activeConditions = new ArrayList<String>();
var activeParams = new ArrayList<Object>();
activeConditions.add("status = 'RUNNING'");
addScopeFilters(routeId, agentIds, activeConditions, activeParams);
String activeWhere = " WHERE " + String.join(" AND ", activeConditions);
Long activeCount = jdbcTemplate.queryForObject(
"SELECT count() FROM route_executions" + activeWhere,
Long.class, activeParams.toArray());
// Previous period (same window shifted back 24h) — read from rollup
Duration window = Duration.between(from, to);
Instant prevFrom = from.minus(Duration.ofHours(24));
Instant prevTo = prevFrom.plus(window);
var prevParams = new ArrayList<Object>();
var prevConditions = new ArrayList<String>();
prevConditions.add("start_time >= ?");
var prevParams = new ArrayList<Object>();
prevConditions.add("bucket >= toStartOfFiveMinutes(?)");
prevParams.add(Timestamp.from(prevFrom));
prevConditions.add("start_time <= ?");
prevConditions.add("bucket <= ?");
prevParams.add(Timestamp.from(prevTo));
addScopeFilters(routeId, agentIds, prevConditions, prevParams);
String prevWhere = " WHERE " + String.join(" AND ", prevConditions);
String prevAggregateSql = "SELECT count() AS total_count, " +
"countIf(status = 'FAILED') AS failed_count, " +
"toInt64(ifNotFinite(avg(duration_ms), 0)) AS avg_duration_ms, " +
"toInt64(ifNotFinite(quantile(0.99)(duration_ms), 0)) AS p99_duration_ms, " +
"countIf(status = 'RUNNING') AS active_count " +
"FROM route_executions" + prevWhere;
String prevRollupSql = "SELECT " +
"countMerge(total_count) AS total_count, " +
"countIfMerge(failed_count) AS failed_count, " +
"toInt64(ifNotFinite(sumMerge(duration_sum) / countMerge(total_count), 0)) AS avg_duration_ms, " +
"toInt64(ifNotFinite(quantileTDigestMerge(0.99)(p99_duration), 0)) AS p99_duration_ms " +
"FROM route_execution_stats_5m" + prevWhere;
PeriodStats prev = jdbcTemplate.queryForObject(prevAggregateSql,
PeriodStats prev = jdbcTemplate.queryForObject(prevRollupSql,
(rs, rowNum) -> new PeriodStats(
rs.getLong("total_count"),
rs.getLong("failed_count"),
rs.getLong("avg_duration_ms"),
rs.getLong("p99_duration_ms"),
rs.getLong("active_count")),
rs.getLong("p99_duration_ms")),
prevParams.toArray());
// Today total (midnight UTC to now) with same scope
// Today total (midnight UTC to now) — read from rollup with same scope
Instant todayStart = Instant.now().truncatedTo(java.time.temporal.ChronoUnit.DAYS);
var todayConditions = new ArrayList<String>();
var todayParams = new ArrayList<Object>();
todayConditions.add("start_time >= ?");
todayConditions.add("bucket >= toStartOfFiveMinutes(?)");
todayParams.add(Timestamp.from(todayStart));
addScopeFilters(routeId, agentIds, todayConditions, todayParams);
String todayWhere = " WHERE " + String.join(" AND ", todayConditions);
Long totalToday = jdbcTemplate.queryForObject(
"SELECT count() FROM route_executions" + todayWhere,
"SELECT countMerge(total_count) FROM route_execution_stats_5m" + todayWhere,
Long.class, todayParams.toArray());
return new ExecutionStats(
current.totalCount, current.failedCount, current.avgDurationMs,
current.p99LatencyMs, current.activeCount,
current.p99LatencyMs, activeCount != null ? activeCount : 0L,
totalToday != null ? totalToday : 0L,
prev.totalCount, prev.failedCount, prev.avgDurationMs, prev.p99LatencyMs);
}
@@ -186,34 +194,32 @@ public class ClickHouseSearchEngine implements SearchEngine {
var conditions = new ArrayList<String>();
var params = new ArrayList<Object>();
conditions.add("start_time >= ?");
conditions.add("bucket >= toStartOfFiveMinutes(?)");
params.add(Timestamp.from(from));
conditions.add("start_time <= ?");
conditions.add("bucket <= ?");
params.add(Timestamp.from(to));
addScopeFilters(routeId, agentIds, conditions, params);
String where = " WHERE " + String.join(" AND ", conditions);
// Use epoch-based bucketing for DateTime64 compatibility
// Re-aggregate 5-minute rollup buckets into the requested interval
String sql = "SELECT " +
"toDateTime(intDiv(toUInt32(toDateTime(start_time)), " + intervalSeconds + ") * " + intervalSeconds + ") AS bucket, " +
"count() AS total_count, " +
"countIf(status = 'FAILED') AS failed_count, " +
"toInt64(ifNotFinite(avg(duration_ms), 0)) AS avg_duration_ms, " +
"toInt64(ifNotFinite(quantile(0.99)(duration_ms), 0)) AS p99_duration_ms, " +
"countIf(status = 'RUNNING') AS active_count " +
"FROM route_executions" + where +
" GROUP BY bucket " +
"ORDER BY bucket";
"toDateTime(intDiv(toUInt32(bucket), " + intervalSeconds + ") * " + intervalSeconds + ") AS ts_bucket, " +
"countMerge(total_count) AS total_count, " +
"countIfMerge(failed_count) AS failed_count, " +
"toInt64(ifNotFinite(sumMerge(duration_sum) / countMerge(total_count), 0)) AS avg_duration_ms, " +
"toInt64(ifNotFinite(quantileTDigestMerge(0.99)(p99_duration), 0)) AS p99_duration_ms " +
"FROM route_execution_stats_5m" + where +
" GROUP BY ts_bucket ORDER BY ts_bucket";
List<StatsTimeseries.TimeseriesBucket> buckets = jdbcTemplate.query(sql, (rs, rowNum) ->
new StatsTimeseries.TimeseriesBucket(
rs.getTimestamp("bucket").toInstant(),
rs.getTimestamp("ts_bucket").toInstant(),
rs.getLong("total_count"),
rs.getLong("failed_count"),
rs.getLong("avg_duration_ms"),
rs.getLong("p99_duration_ms"),
rs.getLong("active_count")
0L
),
params.toArray());

View File

@@ -0,0 +1,31 @@
-- Pre-aggregated 5-minute stats rollup for route executions.
-- Uses AggregatingMergeTree with -State/-Merge combinators so intermediate
-- aggregates can be merged across arbitrary time windows and dimensions.
CREATE TABLE IF NOT EXISTS route_execution_stats_5m (
bucket DateTime('UTC'),
route_id LowCardinality(String),
agent_id LowCardinality(String),
total_count AggregateFunction(count, UInt64),
failed_count AggregateFunction(countIf, UInt8),
duration_sum AggregateFunction(sum, UInt64),
p99_duration AggregateFunction(quantileTDigest(0.99), UInt64)
)
ENGINE = AggregatingMergeTree()
PARTITION BY toYYYYMMDD(bucket)
ORDER BY (agent_id, route_id, bucket)
TTL bucket + toIntervalDay(30)
SETTINGS ttl_only_drop_parts = 1;
CREATE MATERIALIZED VIEW IF NOT EXISTS route_execution_stats_5m_mv
TO route_execution_stats_5m
AS SELECT
toStartOfFiveMinutes(start_time) AS bucket,
route_id,
agent_id,
countState() AS total_count,
countIfState(status = 'FAILED') AS failed_count,
sumState(duration_ms) AS duration_sum,
quantileTDigestState(0.99)(duration_ms) AS p99_duration
FROM route_executions
GROUP BY bucket, route_id, agent_id;

View File

@@ -0,0 +1,16 @@
-- One-time idempotent backfill of existing route_executions into the
-- 5-minute stats rollup table. Safe for repeated execution — the WHERE
-- clause skips the INSERT if the target table already contains data.
INSERT INTO route_execution_stats_5m
SELECT
toStartOfFiveMinutes(start_time) AS bucket,
route_id,
agent_id,
countState() AS total_count,
countIfState(status = 'FAILED') AS failed_count,
sumState(duration_ms) AS duration_sum,
quantileTDigestState(0.99)(duration_ms) AS p99_duration
FROM route_executions
WHERE (SELECT count() FROM route_execution_stats_5m) = 0
GROUP BY bucket, route_id, agent_id;