feat(ci): add long-range metrics benchmark on ECS (#9218)

* feat(ci): add long-range metrics benchmark on ECS

Signed-off-by: WenyXu <wenymedia@gmail.com>

* feat(ci): run warm and lukewarm long-range benchmarks

Signed-off-by: WenyXu <wenymedia@gmail.com>

* fix(ci): simplify long-range inputs and increase disk budget

Signed-off-by: WenyXu <wenymedia@gmail.com>

* ci: run only warm long-range benchmarks

Signed-off-by: WenyXu <wenymedia@gmail.com>

---------

Signed-off-by: WenyXu <wenymedia@gmail.com>
This commit is contained in:
Weny Xu
2026-09-17 14:50:35 +00:00
committed by GitHub
parent 5180538b03
commit 20d87cff48
4 changed files with 342 additions and 1 deletions
@@ -38,6 +38,10 @@ inputs:
description: Start image-provided Docker and grant the runner Docker group access (root-equivalent)
required: false
default: 'false'
enable-host-cache-clear:
description: Allow only host OS cache clearing via sudo on a dedicated benchmark runner
required: false
default: 'false'
resource-group-id:
description: Optional ECS resource group
required: false
@@ -78,6 +82,7 @@ runs:
ALIYUN_ECS_SYSTEM_DISK_GIB: ${{ inputs.system-disk-gib }}
ALIYUN_ECS_TTL_HOURS: ${{ inputs.ttl-hours }}
ALIYUN_ECS_ENABLE_DOCKER: ${{ inputs.enable-docker }}
ALIYUN_ECS_ENABLE_HOST_CACHE_CLEAR: ${{ inputs.enable-host-cache-clear }}
ALIYUN_ECS_RESOURCE_GROUP_ID: ${{ inputs.resource-group-id }}
QUERY_REGRESSION_RUNNER_UID: ${{ inputs.runner-uid }}
QUERY_REGRESSION_RUNNER_GID: ${{ inputs.runner-gid }}
+26 -1
View File
@@ -104,6 +104,7 @@ class ProvisionConfig:
system_disk_gib: int = SYSTEM_DISK_GIB
ttl_hours: int | None = None
enable_docker: bool = False
enable_host_cache_clear: bool = False
def __post_init__(self) -> None:
if not 20 <= self.system_disk_gib <= 2048:
@@ -128,6 +129,7 @@ def render_user_data(
runner_uid: str = "1001",
runner_gid: str = "1001",
enable_docker: bool = False,
enable_host_cache_clear: bool = False,
) -> str:
"""Render the cloud-init shell script for the runner instance."""
docker_setup = ""
@@ -140,6 +142,23 @@ runner_user=$(id -nu {runner_uid})
usermod -aG docker "$runner_user"
runuser -u "$runner_user" -- docker info
'''
cache_clear_setup = ""
if enable_host_cache_clear:
if not runner_uid.isascii() or not runner_uid.isdigit():
raise ValueError("runner-uid must be numeric for host cache clearing")
cache_clear_setup = f'''# Opt-in host-wide cache control for dedicated benchmark instances only.
command -v sudo
command -v visudo
tee_path=$(command -v tee)
[[ "$tee_path" == /usr/bin/tee || "$tee_path" == /bin/tee ]]
runner_user=$(id -nu {runner_uid})
install -d -m 0755 /etc/sudoers.d
printf '#{runner_uid} ALL=(root) NOPASSWD: %s /proc/sys/vm/drop_caches\\n' "$tee_path" > /etc/sudoers.d/o11ybench-cache-clear
chmod 0440 /etc/sudoers.d/o11ybench-cache-clear
visudo -cf /etc/sudoers.d/o11ybench-cache-clear
runuser -u "$runner_user" -- sudo -n -l -- tee /proc/sys/vm/drop_caches
'''
destinations = " ".join(f'"{dst}"' for dst in CACHE_PATHS)
cache_setup = f"""# Caches live on the system disk, are deleted with the instance, and every
@@ -180,7 +199,7 @@ set -euo pipefail
{swap_setup}
{docker_setup}cat > /etc/ephemeral-github-runner.env <<'ENVEOF'
{docker_setup}{cache_clear_setup}cat > /etc/ephemeral-github-runner.env <<'ENVEOF'
RUNNER_NAME={runner_name}
RUNNER_LABELS={runner_label}
RUNNER_TOKEN={runner_token}
@@ -435,6 +454,7 @@ def provision(config: ProvisionConfig) -> int:
config.runner_uid,
config.runner_gid,
enable_docker=config.enable_docker,
enable_host_cache_clear=config.enable_host_cache_clear,
)
)
@@ -498,6 +518,10 @@ def main() -> int:
"--enable-docker", choices=("true", "false"),
default=os.environ.get("ALIYUN_ECS_ENABLE_DOCKER", "false"),
)
parser.add_argument(
"--enable-host-cache-clear", choices=("true", "false"),
default=os.environ.get("ALIYUN_ECS_ENABLE_HOST_CACHE_CLEAR", "false"),
)
parser.add_argument("--repo", default=os.environ.get("GITHUB_REPOSITORY"))
parser.add_argument("--run-id", default=os.environ.get("GITHUB_RUN_ID"))
parser.add_argument("--github-token", default=os.environ.get("GH_PERSONAL_ACCESS_TOKEN"))
@@ -531,6 +555,7 @@ def main() -> int:
system_disk_gib=args.system_disk_gib,
ttl_hours=args.ttl_hours,
enable_docker=args.enable_docker == "true",
enable_host_cache_clear=args.enable_host_cache_clear == "true",
)
return provision(config)
+280
View File
@@ -0,0 +1,280 @@
name: VMBench Long Range
on:
workflow_dispatch:
inputs:
run_greptimedb:
description: Run GreptimeDB
type: boolean
default: true
required: false
run_victoriametrics:
description: Run VictoriaMetrics
type: boolean
default: false
required: false
range_preset:
description: Long-range query window (fixed dataset per preset)
type: choice
options:
- 1w
- 1mo
default: 1w
required: true
greptimedb_tag:
description: Docker Hub greptime/greptimedb tag
type: string
default: latest
required: true
victoriametrics_tag:
description: Docker Hub victoriametrics/victoria-metrics tag or tag@sha256 digest
type: string
default: v1.114.0@sha256:227ad53fd57a4b4430624898a1be0cac39431e1235fa0dc92f567cce38969344
required: true
runtime_image:
description: Aliyun runtime image (tag or digest reference)
type: string
default: greptime-registry.cn-hangzhou.cr.aliyuncs.com/tools/o11ybench-runtime:20260917-65906ca9@sha256:5af45e298019474d4c5089d63031a15f9bceb663890dd87fa0a56876c99c8597
required: true
ecs_instance_type:
description: ECS instance type (independent of Query Regression)
type: string
default: ecs.c9i.2xlarge
required: true
system_disk_gib:
description: System disk GiB (20..2048)
type: number
default: 80
required: true
benchmark_timeout_minutes:
description: Benchmark job timeout in minutes (1..360)
type: number
default: 360
required: true
janitor_ttl_hours:
description: Instance lifetime before janitor eligibility (1..168 hours; at least timeout + 2h)
type: number
default: 8
required: true
db_cpus:
description: CPU limit per DB container
type: string
default: '8'
required: true
db_memory:
description: Memory limit per DB container
type: string
default: 16g
required: true
permissions:
contents: read
env:
RANGE_PRESET: ${{ inputs.range_preset }}
GREPTIMEDB_TAG: ${{ inputs.greptimedb_tag }}
VICTORIAMETRICS_TAG: ${{ inputs.victoriametrics_tag }}
RUNTIME_IMAGE: ${{ inputs.runtime_image }}
ECS_INSTANCE_TYPE: ${{ inputs.ecs_instance_type }}
SYSTEM_DISK_GIB: ${{ inputs.system_disk_gib }}
BENCHMARK_TIMEOUT_MINUTES: ${{ inputs.benchmark_timeout_minutes }}
JANITOR_TTL_HOURS: ${{ inputs.janitor_ttl_hours }}
DB_CPUS: ${{ inputs.db_cpus }}
DB_MEMORY: ${{ inputs.db_memory }}
jobs:
validate:
runs-on: ubuntu-latest
timeout-minutes: 15
outputs:
system_disk_gib: ${{ steps.plan.outputs.system_disk_gib }}
runtime_image: ${{ steps.plan.outputs.runtime_image }}
o11ybench_ref: ${{ steps.plan.outputs.o11ybench_ref }}
run_id: ${{ steps.plan.outputs.run_id }}
plan_artifact: ${{ steps.plan.outputs.plan_artifact }}
steps:
- uses: actions/checkout@v4
with:
path: greptimedb
persist-credentials: false
- name: Test existing ECS lifecycle helpers
run: python3 greptimedb/tests/perf/test_aliyun_ecs_runner_scripts.py
- name: Validate budgets and build warm plan
id: plan
env:
RUN_GREPTIMEDB: ${{ inputs.run_greptimedb }}
RUN_VICTORIAMETRICS: ${{ inputs.run_victoriametrics }}
run: |
set -euo pipefail
[[ -n "$ECS_INSTANCE_TYPE" ]]
for value in "$SYSTEM_DISK_GIB" "$BENCHMARK_TIMEOUT_MINUTES" "$JANITOR_TTL_HOURS"; do
[[ "$value" =~ ^[1-9][0-9]{0,3}$ ]] || { echo 'Resource budgets must be positive integers'; exit 1; }
done
((SYSTEM_DISK_GIB >= 20 && SYSTEM_DISK_GIB <= 2048))
((BENCHMARK_TIMEOUT_MINUTES <= 360 && JANITOR_TTL_HOURS <= 168))
# Match the existing ECS workflow's provision, teardown and queue headroom.
((JANITOR_TTL_HOURS * 60 >= BENCHMARK_TIMEOUT_MINUTES + 120))
[[ "$RUNTIME_IMAGE" == *.cr.aliyuncs.com/* ]]
docker pull "$RUNTIME_IMAGE"
metadata=$(docker image inspect "$RUNTIME_IMAGE")
resolved=$(jq -er '.[0].RepoDigests[0]' <<< "$metadata")
revision=$(jq -er '.[0].Config.Labels["org.opencontainers.image.revision"]' <<< "$metadata")
[[ "$revision" =~ ^[0-9a-f]{40}$ ]]
printf 'runtime_image=%s\no11ybench_ref=%s\nsystem_disk_gib=%s\n' "$resolved" "$revision" "$SYSTEM_DISK_GIB" >> "$GITHUB_OUTPUT"
printf 'run_id=long-range-%s-%s\nplan_artifact=vmbench-plan-%s-%s\n' "$GITHUB_RUN_ID" "$GITHUB_RUN_ATTEMPT" "$GITHUB_RUN_ID" "$GITHUB_RUN_ATTEMPT" >> "$GITHUB_OUTPUT"
args=(--workload vmbench-long-range --range-preset "$RANGE_PRESET"
--db-cpus "$DB_CPUS" --db-memory "$DB_MEMORY"
--run-classification controlled_server_benchmark --no-open-browser)
# Tags become target-spec fields; reject separators before passing them to the planner.
for target in greptimedb victoriametrics; do
case "$target" in
greptimedb) enabled=$RUN_GREPTIMEDB; tag=$GREPTIMEDB_TAG; repo=greptime/greptimedb; url=http://127.0.0.1:14000/v1/prometheus/api/v1/query_range ;;
victoriametrics) enabled=$RUN_VICTORIAMETRICS; tag=$VICTORIAMETRICS_TAG; repo=victoriametrics/victoria-metrics; url=http://127.0.0.1:18428/api/v1/query_range ;;
esac
[[ "$enabled" == true ]] || continue
[[ "$tag" =~ ^[a-zA-Z0-9_][a-zA-Z0-9_.-]{0,127}(@sha256:[0-9a-f]{64})?$ ]]
args+=(--target "$target:name=$target,image=$repo:$tag,url=$url")
done
# The packaged planner owns profile, target and resource-limit validation.
mkdir -p "$RUNNER_TEMP/vmbench-plans"
# Keep the original warm defaults: serial 5/1 and c1/c4 30s/5s.
docker run --rm "$resolved" scripts/o11ybench_plan.py "${args[@]}" \
--run-id "long-range-$GITHUB_RUN_ID-$GITHUB_RUN_ATTEMPT-warm" \
--output-root "/tmp/o11ybench-long-range-$GITHUB_RUN_ID-$GITHUB_RUN_ATTEMPT-warm" \
--cache-policy warm > "$RUNNER_TEMP/vmbench-plans/warm.json"
- uses: actions/upload-artifact@v4
with:
name: vmbench-plan-${{ github.run_id }}-${{ github.run_attempt }}
path: ${{ runner.temp }}/vmbench-plans/
if-no-files-found: error
retention-days: 1
provision:
needs: validate
runs-on: ubuntu-latest
timeout-minutes: 45
outputs:
label: ${{ steps.ecs.outputs.label }}
instance_id: ${{ steps.ecs.outputs.instance_id }}
runner_name: ${{ steps.ecs.outputs.runner_name }}
steps:
- uses: actions/checkout@v4
with:
path: greptimedb
persist-credentials: false
- name: Create ephemeral ECS runner
id: ecs
uses: ./greptimedb/.github/actions/aliyun-ecs-create
with:
access-key-id: ${{ secrets.ALICLOUD_ECS_ACCESS_KEY_ID }}
access-key-secret: ${{ secrets.ALICLOUD_ECS_ACCESS_KEY_SECRET }}
github-token: ${{ secrets.GH_PERSONAL_ACCESS_TOKEN }}
region-id: ${{ vars.ALIYUN_ECS_REGION_ID }}
vswitch-id: ${{ vars.ALIYUN_ECS_VSWITCH_ID }}
security-group-id: ${{ vars.ALIYUN_ECS_SECURITY_GROUP_ID }}
image-id: ${{ vars.QUERY_REGRESSION_ECS_IMAGE_ID }}
instance-type: ${{ env.ECS_INSTANCE_TYPE }}
enable-docker: 'true'
system-disk-gib: ${{ needs.validate.outputs.system_disk_gib }}
ttl-hours: ${{ inputs.janitor_ttl_hours }}
resource-group-id: ${{ vars.ALIYUN_ECS_RESOURCE_GROUP_ID }}
runner-uid: ${{ vars.QUERY_REGRESSION_RUNNER_UID || '1001' }}
runner-gid: ${{ vars.QUERY_REGRESSION_RUNNER_GID || '1001' }}
benchmark:
needs:
- validate
- provision
runs-on: ${{ needs.provision.outputs.label }}
timeout-minutes: ${{ fromJSON(format('{0}', inputs.benchmark_timeout_minutes)) }}
env:
RESOLVED_RUNTIME_IMAGE: ${{ needs.validate.outputs.runtime_image }}
RESOLVED_O11YBENCH_REF: ${{ needs.validate.outputs.o11ybench_ref }}
RUN_ID: ${{ needs.validate.outputs.run_id }}
ARTIFACT_ROOT: /tmp/o11ybench-${{ needs.validate.outputs.run_id }}
METADATA_ROOT: ${{ github.workspace }}/vmbench-metadata
O11YBENCH_DIR: ${{ github.workspace }}/o11ybench
TSDG_PARALLELISM: '8'
steps:
- name: Checkout matching benchmark tools
uses: actions/checkout@v4
with:
repository: GreptimeTeam/o11ybench
token: ${{ secrets.GH_PERSONAL_ACCESS_TOKEN }}
ref: ${{ env.RESOLVED_O11YBENCH_REF }}
path: o11ybench
persist-credentials: false
- uses: actions/download-artifact@v4
with:
name: ${{ needs.validate.outputs.plan_artifact }}
path: ${{ env.METADATA_ROOT }}
- name: Check runner and resolve database images
run: |
set -euo pipefail
[[ "$(uname -m)" == x86_64 ]]
test "$(git -C "$O11YBENCH_DIR" rev-parse HEAD)" = "$RESOLVED_O11YBENCH_REF"
jq --version
docker info
df -h /tmp /var/lib/docker
# Fail before generation if either dataset space or Docker storage is too small.
for path in /tmp /var/lib/docker; do
available=$(df -Pk "$path" | awk 'NR == 2 {print $4}')
((available >= 10 * 1024 * 1024)) || { echo "Need at least 10 GiB free on $path"; exit 1; }
done
docker pull "$RESOLVED_RUNTIME_IMAGE"
docker image inspect "$RESOLVED_RUNTIME_IMAGE" > "$METADATA_ROOT/runtime-image.json"
test "$(jq -r '.[0].Config.Labels["org.opencontainers.image.revision"]' "$METADATA_ROOT/runtime-image.json")" = "$RESOLVED_O11YBENCH_REF"
plan="$METADATA_ROOT/warm.json"
while IFS=$'\t' read -r target reference; do
echo "Resolving $target: $reference"
docker pull "$reference"
docker image inspect "$reference" > "$METADATA_ROOT/$target-image.json"
digest=$(jq -er '.[0].RepoDigests[0]' "$METADATA_ROOT/$target-image.json")
jq --arg target "$target" --arg image "$digest" '.targets |= map(if .name == $target then .image = $image else . end)' "$plan" > "$plan.tmp"
mv "$plan.tmp" "$plan"
done < <(jq -r '.targets[] | [.name, .image] | @tsv' "$plan")
- name: Review execution plan
run: |
set -euo pipefail
python3 o11ybench/scripts/run_benchmark_plan.py \
--plan "$METADATA_ROOT/warm.json" --runtime-image "$RESOLVED_RUNTIME_IMAGE" --dry-run \
> "$METADATA_ROOT/warm-execution-plan.json"
cat "$METADATA_ROOT/warm-execution-plan.json"
- name: Warm serial and concurrent benchmarks
run: |
set -euo pipefail
python3 -u o11ybench/scripts/run_benchmark_plan.py \
--plan "$METADATA_ROOT/warm.json" --runtime-image "$RESOLVED_RUNTIME_IMAGE" --execute \
2>&1 | tee "$METADATA_ROOT/warm-execution.log"
python3 o11ybench/scripts/check_run_benchmark_plan.py --artifacts "$ARTIFACT_ROOT-warm"
- name: Clean up owned containers
if: always()
run: |
docker ps -aq --filter "label=o11ybench.run-id=$RUN_ID-warm" | xargs -r docker rm -fv
- name: Upload HTML and results (no dataset or DB files)
if: always()
uses: actions/upload-artifact@v4
with:
name: vmbench-long-range-${{ github.run_id }}-${{ github.run_attempt }}
path: |
${{ env.METADATA_ROOT }}
${{ env.ARTIFACT_ROOT }}-warm
!${{ env.ARTIFACT_ROOT }}-warm/data/*.bin
!${{ env.ARTIFACT_ROOT }}-warm/**/db-data/**
if-no-files-found: warn
retention-days: 14
teardown:
needs:
- provision
- benchmark
if: ${{ always() && needs.provision.outputs.instance_id != '' }}
runs-on: ubuntu-latest
timeout-minutes: 15
steps:
- uses: actions/checkout@v4
with:
path: greptimedb
persist-credentials: false
- name: Delete ECS and unregister runner
uses: ./greptimedb/.github/actions/aliyun-ecs-delete
with:
access-key-id: ${{ secrets.ALICLOUD_ECS_ACCESS_KEY_ID }}
access-key-secret: ${{ secrets.ALICLOUD_ECS_ACCESS_KEY_SECRET }}
github-token: ${{ secrets.GH_PERSONAL_ACCESS_TOKEN }}
region-id: ${{ vars.ALIYUN_ECS_REGION_ID }}
instance-id: ${{ needs.provision.outputs.instance_id }}
runner-name: ${{ needs.provision.outputs.runner_name }}
@@ -21,6 +21,7 @@ import io
import os
import subprocess
import sys
import tempfile
import unittest
from contextlib import redirect_stderr
from datetime import datetime, timedelta, timezone
@@ -257,6 +258,36 @@ runuser() { echo "runuser $*"; }
["usermod -aG docker custom-runner", "runuser -u custom-runner -- docker info"]
if fail == "0" else []))
def test_host_cache_clear_is_opt_in_and_command_scoped(self) -> None:
args = ("runner", "label", "token", "owner/repo")
self.assertNotIn("o11ybench-cache-clear", provision.render_user_data(*args))
enabled = provision.render_user_data(*args, runner_uid="2001", enable_host_cache_clear=True)
setup = enabled[enabled.index("# Opt-in host-wide"):enabled.index("cat > /etc/ephemeral-github-runner.env")]
self.assertIn("#2001 ALL=(root) NOPASSWD: %s /proc/sys/vm/drop_caches", setup)
self.assertNotIn("NOPASSWD: ALL", setup)
self.assertIn("visudo -cf", setup)
self.assertLess(enabled.index("visudo -cf"), enabled.index("systemctl restart --no-block ephemeral-github-runner.service"))
with self.assertRaisesRegex(ValueError, "runner-uid must be numeric"):
provision.render_user_data(*args, runner_uid="ALL", enable_host_cache_clear=True)
with tempfile.TemporaryDirectory() as tmp:
# Run the actual setup in an isolated path, mocking only privileged tools.
script = setup.replace("/etc/sudoers.d", tmp)
mocks = '''
sudo() { :; }
visudo() { echo "visudo $*"; return "$FAIL_VALIDATE"; }
id() { [[ "$*" == "-nu 2001" ]] || return 1; echo custom-runner; }
runuser() { echo "runuser $*"; }
'''
for fail in ("0", "1"):
result = subprocess.run(["bash", "-euc", mocks + script],
env=os.environ | {"FAIL_VALIDATE": fail}, capture_output=True, text=True)
self.assertEqual(result.returncode, int(fail), result.stderr)
policy = Path(tmp) / "o11ybench-cache-clear"
self.assertEqual(policy.read_text(), "#2001 ALL=(root) NOPASSWD: /usr/bin/tee /proc/sys/vm/drop_caches\n")
self.assertEqual(policy.stat().st_mode & 0o777, 0o440)
self.assertEqual('runuser -u custom-runner -- sudo -n -l -- tee /proc/sys/vm/drop_caches' in result.stdout, fail == "0")
policy.chmod(0o600)
def test_encode_user_data_round_trips(self) -> None:
script = self.render()
self.assertEqual(