Profilage des allocations

<clickhouse>
    <jemalloc_enable_global_profiler>1</jemalloc_enable_global_profiler>
</clickhouse>

<clickhouse>
    <jemalloc_collect_global_profile_samples_in_trace_log>1</jemalloc_collect_global_profile_samples_in_trace_log>
</clickhouse>

SELECT *
FROM numbers(1000000)
ORDER BY number DESC
SETTINGS max_bytes_ratio_before_external_sort = 0
FORMAT `Null`
SETTINGS jemalloc_enable_profiler = 1, jemalloc_collect_profile_samples_in_trace_log = 1

Query id: 8678d8fe-62c5-48b8-b0cd-26851c62dd75

Ok.

0 rows in set. Elapsed: 0.009 sec. Processed 1.00 million rows, 8.00 MB (108.58 million rows/s., 868.61 MB/s.)
Peak memory usage: 12.65 MiB.

SYSTEM FLUSH LOGS trace_log

WITH per_bucket AS
(
    SELECT
        event_time_microseconds AS bucket_time,
        sum(size) AS bucket_sum
    FROM system.trace_log
    WHERE trace_type = 'JemallocSample'
      AND query_id = '8678d8fe-62c5-48b8-b0cd-26851c62dd75'
    GROUP BY bucket_time
)
SELECT
    bucket_time,
    sum(bucket_sum) OVER (
        ORDER BY bucket_time ASC
        ROWS BETWEEN UNBOUNDED PRECEDING AND CURRENT ROW
    ) AS cumulative_size,
    formatReadableSize(cumulative_size) AS cumulative_size_readable
FROM per_bucket
ORDER BY bucket_time

SELECT
    argMax(bucket_time, cumulative_size),
    max(cumulative_size)
FROM
(
    WITH per_bucket AS
    (
        SELECT
            event_time_microseconds AS bucket_time,
            sum(size) AS bucket_sum
        FROM system.trace_log
        WHERE trace_type = 'JemallocSample'
          AND query_id = '8678d8fe-62c5-48b8-b0cd-26851c62dd75'
        GROUP BY bucket_time
    )
    SELECT
        bucket_time,
        sum(bucket_sum) OVER (
            ORDER BY bucket_time ASC
            ROWS BETWEEN UNBOUNDED PRECEDING AND CURRENT ROW
        ) AS cumulative_size,
        formatReadableSize(cumulative_size) AS cumulative_size_readable
    FROM per_bucket
    ORDER BY bucket_time
)

SELECT
    concat(
        '\n',
        arrayStringConcat(
            arrayMap(
                (x, y) -> concat(x, ': ', y),
                arrayMap(x -> addressToLine(x), allocation_trace),
                arrayMap(x -> demangle(addressToSymbol(x)), allocation_trace)
            ),
            '\n'
        )
    ) AS symbolized_trace,
    sum(s) AS per_trace_sum
FROM
(
    SELECT
        ptr,
        sum(size) AS s,
        argMax(trace, event_time_microseconds) AS allocation_trace
    FROM system.trace_log
    WHERE trace_type = 'JemallocSample'
      AND query_id = '8678d8fe-62c5-48b8-b0cd-26851c62dd75'
      AND event_time_microseconds <= '2025-09-04 11:56:21.737139'
    GROUP BY ptr
    HAVING s > 0
)
GROUP BY ALL
ORDER BY per_trace_sum ASC

SELECT * FROM system.jemalloc_profile_text

SELECT * FROM system.jemalloc_profile_text
SETTINGS jemalloc_profile_text_output_format = 'raw'

SELECT * FROM system.jemalloc_profile_text
SETTINGS jemalloc_profile_text_output_format = 'symbolized'

clickhouse-client -q "SELECT * FROM system.jemalloc_profile_text" | flamegraph.pl --color=mem --title="Allocation Flame Graph" --width 2400 > result.svg

clickhouse-client -q "SELECT * FROM system.jemalloc_profile_text SETTINGS jemalloc_profile_text_collapsed_use_count = 1" | flamegraph.pl --color=mem --title="Allocation Count Flame Graph" --width 2400 > result.svg

WITH 60e9 AS min_age_ns
SELECT
    b.arena,
    b.size AS reg_size,
    b.waste AS class_waste,
    s.est_old_objects,
    s.stack
FROM system.jemalloc_arena_bins AS b
INNER JOIN
(
    SELECT
        arena,
        size_class,
        sum(weight) AS est_old_objects,
        arrayStringConcat(arrayMap(a -> demangle(addressToSymbol(a)), arrayReverse(trace)), ';') AS stack
    FROM system.jemalloc_sampled_allocations
    WHERE age_ns > min_age_ns
    GROUP BY arena, size_class, trace
) AS s ON s.size_class = b.index AND s.arena = b.arena
WHERE b.large = 0 AND b.purpose = '' AND b.waste > 1048576
ORDER BY b.waste DESC, s.est_old_objects DESC
SETTINGS allow_introspection_functions=1

WITH 60e9 AS min_age_ns
SELECT format('{} {}',
    if(s.est_old_objects > 0, s.stack, format('[unattributed];arena_{}_class_{}', toString(b.arena), toString(b.size))),
    toString(toUInt64(if(s.est_old_objects > 0, b.waste * s.est_old_objects / sum(s.est_old_objects) OVER (PARTITION BY b.arena, b.index), b.waste))))
FROM system.jemalloc_arena_bins AS b
LEFT JOIN
(
    SELECT
        arena,
        size_class,
        sum(weight) AS est_old_objects,
        arrayStringConcat(arrayMap(a -> demangle(addressToSymbol(a)), arrayReverse(trace)), ';') AS stack
    FROM system.jemalloc_sampled_allocations
    WHERE age_ns > min_age_ns
    GROUP BY arena, size_class, trace
) AS s ON s.size_class = b.index AND s.arena = b.arena
WHERE b.large = 0 AND b.purpose = '' AND b.waste > 0
SETTINGS allow_introspection_functions=1

clickhouse-client -q "..." | flamegraph.pl --color=mem --title="Fragmentation Flame Graph" --width 2400 > result.svg

MALLOC_CONF=prof_prefix:/data/my_current_profile

SYSTEM JEMALLOC FLUSH PROFILE

jeprof /tmp/jemalloc_clickhouse.12345.0.heap.symbolized --output_format [ > output_file]

jeprof path/to/clickhouse path/to/heap/profile --output_format [ > output_file]

jeprof --base /path/to/first.heap.symbolized /path/to/second.heap.symbolized --output_format [ > output_file]

jeprof /tmp/jemalloc_clickhouse.12345.0.heap.symbolized --text > result.txt

jeprof /tmp/jemalloc_clickhouse.12345.0.heap.symbolized --pdf > result.pdf

jeprof /path/to/clickhouse /tmp/jemalloc_clickhouse.12345.0.heap --text > result.txt

jeprof /path/to/clickhouse /tmp/jemalloc_clickhouse.12345.0.heap --pdf > result.pdf

jeprof /tmp/jemalloc_clickhouse.12345.0.heap.symbolized --collapsed > result.collapsed

jeprof /path/to/clickhouse /tmp/jemalloc_clickhouse.12345.0.heap --collapsed > result.collapsed

cat result.collapsed | /path/to/FlameGraph/flamegraph.pl --color=mem --title="Allocation Flame Graph" --width 2400 > result.svg

SELECT *
FROM system.asynchronous_metrics
WHERE metric LIKE '%jemalloc%'
FORMAT Vertical

SELECT * FROM system.jemalloc_stats

echo jmst | nc localhost 9181