From 31d4b4b81c258cd4d3aa3aeebc9bc4f9810e1285 Mon Sep 17 00:00:00 2001
From: Tom Alexander <tom@fizz.buzz>
Date: Mon, 5 May 2025 19:20:01 -0400
Subject: [PATCH] Add some useful promql queries.

---
 google_cloud_logging_queries/kubernetes/oom   |  1 +
 .../get_requests_for_engines                  |  1 +
 .../over_provisioned/cloud_run_job/notes.org  | 21 +++++++++++++++++++
 metrics_explorer/pod_memory_limit/promql      |  1 +
 4 files changed, 24 insertions(+)
 create mode 100644 google_cloud_logging_queries/kubernetes/oom
 create mode 100644 google_cloud_logging_queries/vertex_ai_agent_builder/get_requests_for_engines
 create mode 100644 metrics_explorer/over_provisioned/cloud_run_job/notes.org
 create mode 100644 metrics_explorer/pod_memory_limit/promql

diff --git a/google_cloud_logging_queries/kubernetes/oom b/google_cloud_logging_queries/kubernetes/oom
new file mode 100644
index 0000000..47f9465
--- /dev/null
+++ b/google_cloud_logging_queries/kubernetes/oom
@@ -0,0 +1 @@
+"OOMKilling"
diff --git a/google_cloud_logging_queries/vertex_ai_agent_builder/get_requests_for_engines b/google_cloud_logging_queries/vertex_ai_agent_builder/get_requests_for_engines
new file mode 100644
index 0000000..9f94998
--- /dev/null
+++ b/google_cloud_logging_queries/vertex_ai_agent_builder/get_requests_for_engines
@@ -0,0 +1 @@
+textPayload=~"HTTP Request: POST https://us-discoveryengine.googleapis.com/[^\s/]+/projects/[^\s/]+/locations/us/collections/[^\s/]+/engines/[^\s/]+/"
diff --git a/metrics_explorer/over_provisioned/cloud_run_job/notes.org b/metrics_explorer/over_provisioned/cloud_run_job/notes.org
new file mode 100644
index 0000000..882bd7c
--- /dev/null
+++ b/metrics_explorer/over_provisioned/cloud_run_job/notes.org
@@ -0,0 +1,21 @@
+* Metrics Explorer PromQL Queries
+** Cloud Run Job Memory Utilization
+Line is a percent of memory usage.
+#+begin_src text
+  histogram_quantile(0.99,sum by (job_name,le)(increase(run_googleapis_com:container_memory_utilizations_bucket{monitored_resource="cloud_run_job"}[${__interval}])))
+#+end_src
+** Memory usage bytes
+#+begin_src text
+  histogram_quantile(0.50,sum_over_time(run_googleapis_com:container_memory_usage_bucket{monitored_resource="cloud_run_job"}[${__interval}]))
+#+end_src
+** Memory allocation
+#+begin_src text
+  sum by (job_name)(rate(run_googleapis_com:container_memory_allocation_time{monitored_resource="cloud_run_job"}[${__interval}]))
+#+end_src
+** Difference
+#+begin_src text
+  sum by (job_name)(
+      rate(run_googleapis_com:container_memory_allocation_time{monitored_resource="cloud_run_job"}[${__interval}])
+      - rate(run_googleapis_com:container_memory_usage_bucket{monitored_resource="cloud_run_job"}[${__interval}])
+      )
+#+end_src
diff --git a/metrics_explorer/pod_memory_limit/promql b/metrics_explorer/pod_memory_limit/promql
new file mode 100644
index 0000000..0c8f9d0
--- /dev/null
+++ b/metrics_explorer/pod_memory_limit/promql
@@ -0,0 +1 @@
+sum by (pod_name)(avg_over_time(kubernetes_io:container_memory_limit_bytes{monitored_resource="k8s_container",container_name=~".*airflow-worker.*"}[${__interval}]))