chore: upload system performance test result to Discord (#28458)

upload system performance test result to Discord

system-test/automation_utils.sh

@@ -292,6 +292,52 @@ EOF
   "$SLACK_WEBHOOK_URL"
 }
 
+function upload_results_to_discord() {
+  echo --- Uploading results to Discord Performance Results App
+
+  if [[ -z $DISCORD_WEBHOOK_URL ]] ; then
+    echo "DISCORD_WEBHOOK_URL undefined"
+    exit 1
+  fi
+
+  [[ -n $BUILDKITE_MESSAGE ]] || BUILDKITE_MESSAGE="Message not defined"
+
+  COMMIT=$(git rev-parse HEAD)
+  COMMIT_BUTTON_TEXT="$(echo "$COMMIT" | head -c 8)"
+  COMMIT_URL="https://github.com/solana-labs/solana/commit/${COMMIT}"
+
+  if [[ -n $BUILDKITE_BUILD_URL ]] ; then
+    BUILD_BUTTON_TEXT="Build Kite Job"
+  else
+    BUILD_BUTTON_TEXT="Build URL not defined"
+    BUILDKITE_BUILD_URL="https://buildkite.com/solana-labs/"
+  fi
+
+  GRAFANA_URL="https://internal-metrics.solana.com:3000/d/monitor-${CHANNEL:-edge}/cluster-telemetry-${CHANNEL:-edge}?var-testnet=${TESTNET_TAG:-testnet-automation}&from=${TESTNET_START_UNIX_MSECS:-0}&to=${TESTNET_FINISH_UNIX_MSECS:-0}"
+
+  [[ -n $RESULT_DETAILS ]] || RESULT_DETAILS="Undefined"
+  SANITIZED_RESULT=${RESULT_DETAILS//$'\n'/"\n"}
+
+  [[ -n $TEST_CONFIGURATION ]] || TEST_CONFIGURATION="Undefined"
+
+  curl "$DISCORD_WEBHOOK_URL" \
+      -X POST \
+      -H "Content-Type: application/json" \
+      -d @- <<EOF
+{
+  "username": "System Performance Test",
+  "content": "\
+**$BUILDKITE_MESSAGE**\n\
+[$COMMIT_BUTTON_TEXT](<$COMMIT_URL>) | [$BUILD_BUTTON_TEXT](<$BUILDKITE_BUILD_URL>) | [Grafana](<$GRAFANA_URL>)\n\
+Test Configuration:\n\
+\`\`\`$TEST_CONFIGURATION\`\`\`\n\
+Result Details:\n\
+\`\`\`$SANITIZED_RESULT\`\`\`\n\
+"
+}
+EOF
+}
+
 function get_net_launch_software_version_launch_args() {
   declare channel="${1?}"
   declare artifact_basename="${2?}"

system-test/deprecated-testcases/colo-cpu-only-perf.yml

@@ -3,6 +3,7 @@ steps:
     label: "COLO performance testnet CPU only"
     env:
       UPLOAD_RESULTS_TO_SLACK: "true"
+      UPLOAD_RESULTS_TO_DISCORD: "true"
       CLOUD_PROVIDER: "colo"
       TESTNET_TAG: "colo-perf-cpu-only"
       ENABLE_GPU: "false"

system-test/deprecated-testcases/colo-gpu-perf-high-txcount.yml

@@ -3,6 +3,7 @@ steps:
     label: "COLO performance testnet GPU enabled - High Tx Count"
     env:
       UPLOAD_RESULTS_TO_SLACK: "true"
+      UPLOAD_RESULTS_TO_DISCORD: "true"
       CLOUD_PROVIDER: "colo"
       TESTNET_TAG: "colo-perf-gpu-enabled"
       ENABLE_GPU: "true"

system-test/deprecated-testcases/colo-gpu-perf.yml

@@ -3,6 +3,7 @@ steps:
     label: "COLO performance testnet GPU enabled"
     env:
       UPLOAD_RESULTS_TO_SLACK: "true"
+      UPLOAD_RESULTS_TO_DISCORD: "true"
       CLOUD_PROVIDER: "colo"
       TESTNET_TAG: "colo-perf-gpu-enabled"
       ENABLE_GPU: "true"

system-test/deprecated-testcases/gce-gpu-perf-100-node.yml

@@ -3,6 +3,7 @@ steps:
     label: "GCE - GPU Enabled 100 Nodes"
     env:
       UPLOAD_RESULTS_TO_SLACK: "true"
+      UPLOAD_RESULTS_TO_DISCORD: "true"
       CLOUD_PROVIDER: "gce"
       TESTNET_TAG: "gce-perf-gpu-enabled"
       TEST_DURATION_SECONDS: 600

system-test/partition-testcases/colo-3-partition.yml

@@ -3,6 +3,7 @@ steps:
     label: "Colo - CPU Only - Partial Loss 3 Partitions"
     env:
       UPLOAD_RESULTS_TO_SLACK: "true"
+      UPLOAD_RESULTS_TO_DISCORD: "true"
       CLOUD_PROVIDER: "colo"
       TESTNET_TAG: "colo-perf-cpu-only"
       NUMBER_OF_VALIDATOR_NODES: 2

system-test/partition-testcases/colo-partition-2-1-test.yml

@@ -3,6 +3,7 @@ steps:
     label: "Colo - CPU Only - Complete Loss 2 - 1 Partition"
     env:
       UPLOAD_RESULTS_TO_SLACK: "true"
+      UPLOAD_RESULTS_TO_DISCORD: "true"
       CLOUD_PROVIDER: "colo"
       TESTNET_TAG: "colo-perf-cpu-only"
       NUMBER_OF_VALIDATOR_NODES: 2

system-test/partition-testcases/colo-partition-long-sanity-test.yml

@@ -3,6 +3,7 @@ steps:
     label: "Colo - CPU Only - Partial Loss 3 Partitions"
     env:
       UPLOAD_RESULTS_TO_SLACK: "true"
+      UPLOAD_RESULTS_TO_DISCORD: "true"
       CLOUD_PROVIDER: "colo"
       TESTNET_TAG: "colo-perf-cpu-only"
       NUMBER_OF_VALIDATOR_NODES: 2

system-test/partition-testcases/colo-partition-no-superminority-test.yml

@@ -3,6 +3,7 @@ steps:
     label: "Colo - CPU Only - Complete Loss 4 Partitions"
     env:
       UPLOAD_RESULTS_TO_SLACK: "true"
+      UPLOAD_RESULTS_TO_DISCORD: "true"
       CLOUD_PROVIDER: "colo"
       TESTNET_TAG: "colo-perf-cpu-only"
       NUMBER_OF_VALIDATOR_NODES: 2

system-test/partition-testcases/colo-partition-once-then-stabilize.yml

@@ -3,6 +3,7 @@ steps:
     label: "Colo - CPU Only - 1 minute partition then 5 minute stabilization"
     env:
       UPLOAD_RESULTS_TO_SLACK: "true"
+      UPLOAD_RESULTS_TO_DISCORD: "true"
       CLOUD_PROVIDER: "colo"
       TESTNET_TAG: "colo-perf-cpu-only"
       NUMBER_OF_VALIDATOR_NODES: 2

system-test/partition-testcases/gce-5-node-3-partition.yml

@@ -3,6 +3,7 @@ steps:
     label: "GCE - CPU Only 5 Node - Partial Loss 3 Partitions"
     env:
       UPLOAD_RESULTS_TO_SLACK: "true"
+      UPLOAD_RESULTS_TO_DISCORD: "true"
       CLOUD_PROVIDER: "gce"
       TESTNET_TAG: "gce-perf-cpu-only"
       NUMBER_OF_VALIDATOR_NODES: 5

system-test/partition-testcases/gce-5-node-single-region-2-partitions.yml

@@ -3,6 +3,7 @@ steps:
     label: "GCE - 2 even partitions with full loss"
     env:
       UPLOAD_RESULTS_TO_SLACK: "true"
+      UPLOAD_RESULTS_TO_DISCORD: "true"
       CLOUD_PROVIDER: "gce"
       TESTNET_TAG: "gce-perf-cpu-only"
       NUMBER_OF_VALIDATOR_NODES: 5

system-test/partition-testcases/gce-partition-once-then-stabilize.yml

@@ -3,6 +3,7 @@ steps:
     label: "GCE - CPU Only - 1 minute partition then 5 minute stabilization"
     env:
       UPLOAD_RESULTS_TO_SLACK: "true"
+      UPLOAD_RESULTS_TO_DISCORD: "true"
       CLOUD_PROVIDER: "gce"
       TESTNET_TAG: "gce-perf-cpu-only"
       NUMBER_OF_VALIDATOR_NODES: 5

system-test/partition-testcases/gce-partition-recovery.yml

@@ -3,6 +3,7 @@ steps:
     label: "Partition recovery on GCE"
     env:
       UPLOAD_RESULTS_TO_SLACK: "true"
+      UPLOAD_RESULTS_TO_DISCORD: "true"
       CLOUD_PROVIDER: "gce"
       ENABLE_GPU: "false"
       NUMBER_OF_VALIDATOR_NODES: 9

system-test/partition-testcases/gce-partition-with-offline.yml

@@ -3,6 +3,7 @@ steps:
     label: "GCE - CPU Only 5 Node - 20% network offline with 2 partitions"
     env:
       UPLOAD_RESULTS_TO_SLACK: "true"
+      UPLOAD_RESULTS_TO_DISCORD: "true"
       CLOUD_PROVIDER: "gce"
       TESTNET_TAG: "gce-perf-cpu-only"
       NUMBER_OF_VALIDATOR_NODES: 4

system-test/performance-testcases/aws-cpu-only-perf-10-node.yml

@@ -3,6 +3,7 @@ steps:
     label: "AWS - CPU Only 10 Node"
     env:
       UPLOAD_RESULTS_TO_SLACK: "true"
+      UPLOAD_RESULTS_TO_DISCORD: "true"
       CLOUD_PROVIDER: "ec2"
       TESTNET_TAG: "aws-perf-cpu-only"
       TEST_DURATION_SECONDS: 600

system-test/performance-testcases/aws-cpu-only-perf-5-node.yml

@@ -3,6 +3,7 @@ steps:
     label: "AWS - CPU Only 5 Node"
     env:
       UPLOAD_RESULTS_TO_SLACK: "true"
+      UPLOAD_RESULTS_TO_DISCORD: "true"
       CLOUD_PROVIDER: "ec2"
       TESTNET_TAG: "aws-perf-cpu-only"
       TEST_DURATION_SECONDS: 600

system-test/performance-testcases/azure-cpu-only-perf-5-node.yml

@@ -3,6 +3,7 @@ steps:
     label: "Azure - CPU Only 5 Node"
     env:
       UPLOAD_RESULTS_TO_SLACK: "true"
+      UPLOAD_RESULTS_TO_DISCORD: "true"
       CLOUD_PROVIDER: "azure"
       TESTNET_TAG: "azure-perf-cpu-only"
       TEST_DURATION_SECONDS: 600

system-test/performance-testcases/colo-cpu-only-perf-4-val-1-client.yml

@@ -3,6 +3,7 @@ steps:
     label: "COLO performance testnet CPU only (reduced validator and client count)"
     env:
       UPLOAD_RESULTS_TO_SLACK: "true"
+      UPLOAD_RESULTS_TO_DISCORD: "true"
       CLOUD_PROVIDER: "colo"
       TESTNET_TAG: "colo-perf-cpu-only"
       ENABLE_GPU: "false"

system-test/performance-testcases/colo-gpu-perf-4-val-1-client.yml

@@ -3,6 +3,7 @@ steps:
     label: "COLO performance testnet GPU enabled (reduced validator and client count)"
     env:
       UPLOAD_RESULTS_TO_SLACK: "true"
+      UPLOAD_RESULTS_TO_DISCORD: "true"
       CLOUD_PROVIDER: "colo"
       TESTNET_TAG: "colo-perf-gpu-enabled"
       ENABLE_GPU: "true"

system-test/performance-testcases/colo-gpu-perf-high-txcount-4-val-1-client.yml

@@ -3,6 +3,7 @@ steps:
     label: "COLO performance testnet GPU enabled - High Tx Count (reduced validator and client count)"
     env:
       UPLOAD_RESULTS_TO_SLACK: "true"
+      UPLOAD_RESULTS_TO_DISCORD: "true"
       CLOUD_PROVIDER: "colo"
       TESTNET_TAG: "colo-perf-gpu-enabled"
       ENABLE_GPU: "true"

system-test/performance-testcases/gce-cpu-only-perf-10-node.yml

@@ -3,6 +3,7 @@ steps:
     label: "GCE performance testnets CPU ONLY"
     env:
       UPLOAD_RESULTS_TO_SLACK: "true"
+      UPLOAD_RESULTS_TO_DISCORD: "true"
       CLOUD_PROVIDER: "gce"
       TESTNET_TAG: "gce-perf-cpu-only"
       TEST_DURATION_SECONDS: 600

system-test/performance-testcases/gce-cpu-only-perf-5-node-single-region.yml

@@ -3,6 +3,7 @@ steps:
     label: "GCE - CPU Only 5 Node Single Zone"
     env:
       UPLOAD_RESULTS_TO_SLACK: "true"
+      UPLOAD_RESULTS_TO_DISCORD: "true"
       CLOUD_PROVIDER: "gce"
       TESTNET_TAG: "gce-perf-cpu-only"
       TEST_DURATION_SECONDS: 600

system-test/performance-testcases/gce-cpu-only-perf-5-node.yml

@@ -3,6 +3,7 @@ steps:
     label: "GCE - CPU Only 5 Node"
     env:
       UPLOAD_RESULTS_TO_SLACK: "true"
+      UPLOAD_RESULTS_TO_DISCORD: "true"
       CLOUD_PROVIDER: "gce"
       TESTNET_TAG: "gce-perf-cpu-only"
       TEST_DURATION_SECONDS: 600

system-test/performance-testcases/gce-gpu-perf-10-node-single-region.yml

@@ -3,6 +3,7 @@ steps:
     label: "GCE - GPU Enabled 10 Nodes"
     env:
       UPLOAD_RESULTS_TO_SLACK: "true"
+      UPLOAD_RESULTS_TO_DISCORD: "true"
       CLOUD_PROVIDER: "gce"
       TESTNET_TAG: "gce-perf-gpu-enabled"
       TEST_DURATION_SECONDS: 600

system-test/performance-testcases/gce-gpu-perf-10-node.yml

@@ -3,6 +3,7 @@ steps:
     label: "GCE - GPU Enabled 10 Nodes"
     env:
       UPLOAD_RESULTS_TO_SLACK: "true"
+      UPLOAD_RESULTS_TO_DISCORD: "true"
       CLOUD_PROVIDER: "gce"
       TESTNET_TAG: "gce-perf-gpu-enabled"
       TEST_DURATION_SECONDS: 600

system-test/performance-testcases/gce-gpu-perf-25-node-single-region.yml

@@ -3,6 +3,7 @@ steps:
     label: "GCE - GPU Enabled 25 Nodes"
     env:
       UPLOAD_RESULTS_TO_SLACK: "true"
+      UPLOAD_RESULTS_TO_DISCORD: "true"
       CLOUD_PROVIDER: "gce"
       TESTNET_TAG: "gce-perf-gpu-enabled"
       TEST_DURATION_SECONDS: 600

system-test/performance-testcases/gce-gpu-perf-25-node.yml

@@ -3,6 +3,7 @@ steps:
     label: "GCE - GPU Enabled 25 Nodes"
     env:
       UPLOAD_RESULTS_TO_SLACK: "true"
+      UPLOAD_RESULTS_TO_DISCORD: "true"
       CLOUD_PROVIDER: "gce"
       TESTNET_TAG: "gce-perf-gpu-enabled"
       TEST_DURATION_SECONDS: 600

system-test/performance-testcases/gce-gpu-perf-5-node-single-region.yml

@@ -3,6 +3,7 @@ steps:
     label: "GCE - GPU Enabled 5 Nodes"
     env:
       UPLOAD_RESULTS_TO_SLACK: "true"
+      UPLOAD_RESULTS_TO_DISCORD: "true"
       CLOUD_PROVIDER: "gce"
       TESTNET_TAG: "gce-perf-gpu-enabled"
       TEST_DURATION_SECONDS: 600

system-test/performance-testcases/gce-gpu-perf-5-node.yml

@@ -3,6 +3,7 @@ steps:
     label: "GCE - GPU Enabled 5 Nodes"
     env:
       UPLOAD_RESULTS_TO_SLACK: "true"
+      UPLOAD_RESULTS_TO_DISCORD: "true"
       CLOUD_PROVIDER: "gce"
       TESTNET_TAG: "gce-perf-gpu-enabled"
       TEST_DURATION_SECONDS: 600

system-test/performance-testcases/gce-gpu-perf-50-node-single-region.yml

@@ -3,6 +3,7 @@ steps:
     label: "GCE - GPU Enabled 50 Nodes"
     env:
       UPLOAD_RESULTS_TO_SLACK: "true"
+      UPLOAD_RESULTS_TO_DISCORD: "true"
       CLOUD_PROVIDER: "gce"
       TESTNET_TAG: "gce-perf-gpu-enabled"
       TEST_DURATION_SECONDS: 600

system-test/performance-testcases/gce-gpu-perf-50-node.yml

@@ -3,6 +3,7 @@ steps:
     label: "GCE - GPU Enabled 50 Nodes"
     env:
       UPLOAD_RESULTS_TO_SLACK: "true"
+      UPLOAD_RESULTS_TO_DISCORD: "true"
       CLOUD_PROVIDER: "gce"
       TESTNET_TAG: "gce-perf-gpu-enabled"
       TEST_DURATION_SECONDS: 600

system-test/sanity-testcases/colo-cpu-only-quick-sanity-test.yml

@@ -3,6 +3,7 @@ steps:
     label: "SANITY - Short duration Colo perf sanity.  1 val, 1 client, CPU-only."
     env:
       UPLOAD_RESULTS_TO_SLACK: "true"
+      UPLOAD_RESULTS_TO_DISCORD: "true"
       CLOUD_PROVIDER: "colo"
       TESTNET_TAG: "colo-perf-cpu-only"
       ENABLE_GPU: "false"

system-test/sanity-testcases/colo-partition-sanity-test.yml

@@ -3,6 +3,7 @@ steps:
     label: "SANITY TEST - Colo - CPU Only - Partial Loss 3 Partitions"
     env:
       UPLOAD_RESULTS_TO_SLACK: "true"
+      UPLOAD_RESULTS_TO_DISCORD: "true"
       CLOUD_PROVIDER: "colo"
       TESTNET_TAG: "colo-perf-cpu-only"
       NUMBER_OF_VALIDATOR_NODES: 2

system-test/stability-testcases/colo-long-duration-cpu-only-perf.yml

@@ -3,6 +3,7 @@ steps:
     label: "COLO 1 hour performance & stability CPU only"
     env:
       UPLOAD_RESULTS_TO_SLACK: "true"
+      UPLOAD_RESULTS_TO_DISCORD: "true"
       CLOUD_PROVIDER: "colo"
       TESTNET_TAG: "colo-perf-cpu-only"
       ENABLE_GPU: "false"

system-test/stability-testcases/colo-long-duration-gpu-perf.yml

@@ -3,6 +3,7 @@ steps:
     label: "COLO 1 hour performance & stability GPU enabled"
     env:
       UPLOAD_RESULTS_TO_SLACK: "true"
+      UPLOAD_RESULTS_TO_DISCORD: "true"
       CLOUD_PROVIDER: "colo"
       TESTNET_TAG: "colo-perf-gpu-enabled"
       ENABLE_GPU: "true"

system-test/stability-testcases/gce-perf-stability-5-node-single-region.yml

@@ -3,6 +3,7 @@ steps:
     label: "GCE - 1 hour perf stability"
     env:
       UPLOAD_RESULTS_TO_SLACK: "true"
+      UPLOAD_RESULTS_TO_DISCORD: "true"
       CLOUD_PROVIDER: "gce"
       TESTNET_TAG: "gce-perf-cpu-only"
       TEST_DURATION_SECONDS: 3600

system-test/stability-testcases/gce-stability-5-node.yml

@@ -3,6 +3,7 @@ steps:
     label: "GCE - 8 hour Stability - 5 Node"
     env:
       UPLOAD_RESULTS_TO_SLACK: "true"
+      UPLOAD_RESULTS_TO_DISCORD: "true"
       CLOUD_PROVIDER: "gce"
       TESTNET_TAG: "gce-stability"
       TEST_DURATION_SECONDS: 28800

system-test/stake-operations-testcases/offline_stake_colo.yml

@@ -3,6 +3,7 @@ steps:
     label: "Running Offline Stake Operations Tests on Colo"
     env:
       UPLOAD_RESULTS_TO_SLACK: "true"
+      UPLOAD_RESULTS_TO_DISCORD: "true"
       CLOUD_PROVIDER: "colo"
       ENABLE_GPU: "false"
       TEST_DURATION_SECONDS: 30

system-test/stake-operations-testcases/offline_stake_gce.yml

@@ -3,6 +3,7 @@ steps:
     label: "Running Offline Stake Operations Tests on GCE"
     env:
       UPLOAD_RESULTS_TO_SLACK: "true"
+      UPLOAD_RESULTS_TO_DISCORD: "true"
       CLOUD_PROVIDER: "gce"
       ENABLE_GPU: "false"
       TEST_DURATION_SECONDS: 30

system-test/testnet-automation.sh

@@ -22,6 +22,9 @@ $*"
     upload_results_to_slack
   fi
 
+  if [[ "$UPLOAD_RESULTS_TO_DISCORD" = "true" ]]; then
+    upload_results_to_discord
+  fi
 
   (
     execution_step "Collecting Logfiles from Nodes"