rust/greptimedb
1
0
Fork 0
mirror of https://github.com/GreptimeTeam/greptimedb.git synced 2026-10-12 23:12:28 +00:00
Code Issues Packages Projects Releases Wiki Activity

test(perf): qualify dictionary group-by cache on shared flat SSTs

Browse Source 
Add an explicit SQL grouping case over two immutable flat SSTs with 2097152 rows, 4096 distinct primary-key series and 16 dc groups. Include Utf8, ungrouped and high-cardinality controls plus diagnostic EXPLAIN samples. The existing generator and runner are unchanged. Result equality and numeric invariants must be audited from raw CI samples; a passing latency threshold is not a speedup claim.

Signed-off-by: discord9 <55937128+discord9@users.noreply.github.com>
This commit is contained in:
discord9
2026-10-08 15:31:10 +08:00
parent c82da8308b
commit cc8b54fd39
1 changed files with 125 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files
tests/perf/query_cases/sql_sum_group_by_dict_tag/case.toml
+125
View File
@@ -0,0 +1,125 @@
# Dictionary-tag SUM GROUP BY on immutable flat SSTs. Both benchmark arms receive
# the same materialized data. The deterministic-wave generator is reproducible,
# but does not reproduce any original counter-value distribution exactly.
# Expected shape: 16 groups x 131072 rows for dc, 4096 groups x 512 rows for
# instance, and 2097152 rows for the ungrouped control. Raw sample rows are
# retained; audit complete normalized results, including sums with an
# appropriate floating-point tolerance, rather than counts alone. Success only
# checks requests and configured thresholds; it is not a speedup claim. The
# workload targets reduced dictionary mapping work at this scale; actual Arc
# reuse and dictionary D/R need runtime proof.
[case]
name = "sql_sum_group_by_dict_tag"
description = "dictionary-tag SUM GROUP BY on immutable flat SSTs"
[scenario]
kind = "direct_readable_sst"
seed = 25652
[[scenario.tables]]
database = "public"
name = "sql_sum_group_by_dict_tag"
engine = "mito"
append_mode = true
sst_format = "flat"
primary_key = ["dc", "device", "instance"]
time_index = "ts"
[[scenario.tables.columns]]
name = "dc"
type = "STRING"
semantic = "tag"
distribution = { kind = "cardinality", values = 16, prefix = "dc" }
[[scenario.tables.columns]]
name = "device"
type = "STRING"
semantic = "tag"
distribution = { kind = "cardinality", values = 4, prefix = "device" }
# instance in the primary key retains 4096 distinct series; dc/device alone
# would represent only 16 actual keys.
[[scenario.tables.columns]]
name = "instance"
type = "STRING"
semantic = "tag"
distribution = { kind = "cardinality", values = 4096, prefix = "instance" }
[[scenario.tables.columns]]
name = "val"
type = "DOUBLE"
semantic = "field"
distribution = { kind = "deterministic_wave", min = 0.0, max = 100.0 }
[[scenario.tables.columns]]
name = "ts"
type = "TIMESTAMP(3)"
semantic = "timestamp"
[scenario.layout]
regions = 1
sst_count = 2
rows_per_sst = 1048576
row_group_size = 8192
series_count = 4096
start_unix_nanos = 1704067200000000000
step_nanos = 15000000000
time_range_layout = "non_overlapping_per_sst"
series_layout = "timestamp_major"
[[scenario.queries]]
name = "group_by_dc"
kind = "sql"
query = "SELECT dc, SUM(val) AS sum_val, COUNT(*) AS row_count FROM sql_sum_group_by_dict_tag GROUP BY dc ORDER BY dc"
warmup = 3
iterations = 20
[scenario.queries.thresholds]
max_candidate_latency_regression_pct = 10
# Intended Utf8 control; verify the actual optimized plan and Arrow arrays at
# runtime rather than assuming the cast survives planning.
[[scenario.queries]]
name = "group_by_dc_utf8_control"
kind = "sql"
query = "SELECT CAST(dc AS VARCHAR) AS dc, SUM(val) AS sum_val, COUNT(*) AS row_count FROM sql_sum_group_by_dict_tag GROUP BY CAST(dc AS VARCHAR) ORDER BY dc"
warmup = 3
iterations = 9
[scenario.queries.thresholds]
max_candidate_latency_regression_pct = 10
[[scenario.queries]]
name = "ungrouped_control"
kind = "sql"
query = "SELECT SUM(val) AS sum_val, COUNT(*) AS row_count FROM sql_sum_group_by_dict_tag"
warmup = 3
iterations = 9
[scenario.queries.thresholds]
max_candidate_latency_regression_pct = 10
[[scenario.queries]]
name = "group_by_instance_high_cardinality"
kind = "sql"
query = "SELECT instance, SUM(val) AS sum_val, COUNT(*) AS row_count FROM sql_sum_group_by_dict_tag GROUP BY instance ORDER BY instance"
warmup = 3
iterations = 9
[scenario.queries.thresholds]
max_candidate_latency_regression_pct = 10
[[scenario.queries]]
name = "explain_group_by_dc"
kind = "sql"
query = "EXPLAIN ANALYZE VERBOSE SELECT dc, SUM(val) AS sum_val, COUNT(*) AS row_count FROM sql_sum_group_by_dict_tag GROUP BY dc ORDER BY dc"
warmup = 0
iterations = 1
[[scenario.queries]]
name = "explain_group_by_dc_utf8_control"
kind = "sql"
query = "EXPLAIN ANALYZE VERBOSE SELECT CAST(dc AS VARCHAR) AS dc, SUM(val) AS sum_val, COUNT(*) AS row_count FROM sql_sum_group_by_dict_tag GROUP BY CAST(dc AS VARCHAR) ORDER BY dc"
warmup = 0
iterations = 1