fix(asap-tools): support Google cluster-data experiments

Author: milindsrivastava1997

asap-tools/experiments/config/config.yaml

@@ -98,7 +98,7 @@ query_engine:
 
 # Controller configuration
 controller:
-  punting: ???  # REQUIRED: Enable query punting based on performance heuristics (should_be_performant check)
+  punting: false  # Enable query punting based on performance heuristics (should_be_performant check)
 
 # Optional manual planner windowing override. Omit this block to preserve the
 # default tumbling-window behavior. For tumbling windows, omit

asap-tools/experiments/config/experiment_type/cluster_data_google.yaml

@@ -9,6 +9,11 @@ experiment:
   - mode: baseline
     server: prometheus
 
+# Monitoring tool configuration
+monitoring:
+  tool: prometheus
+  deployment_mode: bare_metal
+
 servers:
 - name: prometheus
   url: http://localhost:9090
@@ -34,6 +39,9 @@ exporters:
       #parts_mode: "all-parts"  # or "part-index" with part_index: 0
       parts_mode: "part-index"  # or "part-index" with part_index: 0
       part_index: 0
+      # A Google task_usage scrape contains many active task series.
+      scrape_interval: "60s"
+      scrape_timeout: "30s"
       # Optional configuration
       # log_level: "INFO"  # DEBUG, INFO, WARN, ERROR (default: INFO)
       # memory_limit: "2g"

asap-tools/experiments/generate_prometheus_config.py

@@ -282,14 +282,17 @@ def main(args, experiment_config=None):
 
         scrape_config = {
             "job_name": "cluster_data_exporter",
-            "scrape_interval": "10s",  # Fast scrape interval for high-resolution cluster data
+            "scrape_interval": cde_config.get("scrape_interval", "10s"),
             "static_configs": [
                 {
                     "targets": [target],
                 }
             ],
         }
 
+        if "scrape_timeout" in cde_config:
+            scrape_config["scrape_timeout"] = cde_config["scrape_timeout"]
+
         # Add metric_relabel_configs to only keep required metrics
         add_metric_relabel_configs(
             scrape_config, "cluster_data_exporter", experiment_config

docs/03-how-to-guides/operations/run-alibaba-cluster-data-experiment.md

@@ -109,6 +109,87 @@ python3 experiment_run_e2e.py \
 The default run takes about three minutes and tears down the services when
 it finishes. Use `flow.no_teardown=true` only when debugging service state.
 
+## Run a Google Cluster-Data Experiment
+
+The Google trace is already split into compressed `task_usage` parts, so no
+preprocessing is required. The benchmark data is under:
+
+```text
+../../benchmarks/metrics_observability/data/google-cluster-data/ClusterData2011/clusterdata-2011-2/task_usage
+```
+
+The checked-in Google configuration uses `part_index: 0`, which is useful for
+a one-node experiment. The first part is approximately 92 MB compressed and
+381 MB uncompressed. Copy it to node19:
+
+```bash
+GOOGLE_DATA="$PWD/../../benchmarks/metrics_observability/data/google-cluster-data/ClusterData2011/clusterdata-2011-2/task_usage"
+CDE_HOST="node19.scratch2.cloudmigration-PG0.utah.cloudlab.us"
+
+ssh -o StrictHostKeyChecking=no "milindsr@$CDE_HOST" \
+  'mkdir -p /scratch/sketch_db_for_prometheus/cluster_traces'
+
+rsync -ah --info=progress2 -e 'ssh -o StrictHostKeyChecking=no' \
+  "$GOOGLE_DATA/part-00000-of-00500.csv.gz" \
+  "milindsr@$CDE_HOST:/scratch/sketch_db_for_prometheus/cluster_traces/"
+```
+
+Build the same exporter image used for Alibaba if it is not already present:
+
+```bash
+ssh -o StrictHostKeyChecking=no "milindsr@$CDE_HOST" \
+  'cd /scratch/sketch_db_for_prometheus/code/asap-tools/data-sources/prometheus-exporters/cluster_data_exporter &&
+   bash installation/install.sh'
+```
+
+Run the Google experiment from `asap-tools/experiments`:
+
+```bash
+python3 experiment_run_e2e.py \
+  experiment.name=google_cluster_data_offset18 \
+  experiment_type=cluster_data_google \
+  providers.cloudlab.username=milindsr \
+  providers.cloudlab.hostname_suffix=scratch2.cloudmigration-PG0.utah.cloudlab.us \
+  providers.cloudlab.node_offset=18 \
+  providers.cloudlab.num_nodes=1 \
+  cluster_data_directory=/scratch/sketch_db_for_prometheus/cluster_traces
+```
+
+This configuration runs both `sketchdb` and `baseline` modes. Google
+`task_usage` produces a much larger `/metrics` response than the Alibaba
+Node trace, so the Google config sets the cluster-data scrape interval to
+60 seconds and timeout to 30 seconds. The Prometheus config generator honors
+these optional exporter settings; Alibaba retains its 10-second default.
+
+The default query is:
+
+```promql
+quantile by () (0.5, google_mean_cpu_usage_rate_0)
+```
+
+For Google data, the exporter labels are `job_id`, `task_index`, and
+`machine_id` (along with the scrape-added `instance` and `job` labels). To
+change the aggregate column, use the corresponding metric name and update
+the `metrics` entry in
+`asap-tools/experiments/config/experiment_type/cluster_data_google.yaml`.
+For example, `canonical-memory-usage` becomes
+`google_canonical_memory_usage_0`. To group the result by a data label, change
+`by ()` to e.g. `by (machine_id)` and include that label in the `metrics`
+entry's label list.
+
+Because Google has separate baseline and sketchdb modes, the latency wrapper
+can be used after the run:
+
+```bash
+cd /home/milind/Desktop/cmu/research/sketch_db_for_prometheus/code/ASAPQuery/asap-tools/experiments/post_experiment
+PYTHONPATH=/home/milind/Desktop/cmu/research/sketch_db_for_prometheus/code/ASAPQuery/asap-common/dependencies/py/promql_utilities \
+./run_compare_latencies.sh google_cluster_data_offset18
+
+python3 analyze_monitor_output.py \
+  /home/milind/Desktop/cmu/research/sketch_db_for_prometheus/code/ASAPQuery/experiment_outputs/google_cluster_data_offset18/sketchdb/remote_monitor_output/monitor_output.json \
+  --print
+```
+
 ## Change the query's aggregation column or grouping label
 
 The raw 2021 Node CSV has columns like: