mirror of
https://github.com/samber/awesome-prometheus-alerts.git
synced 2026-06-26 03:17:07 +08:00
fix: address review feedback on runtime alerts
- JVM non-heap: guard against unbounded metaspace (max_bytes = -1) - JVM old gen GC: note regex only matches CMS/G1/Parallel collectors - JVM/Python file descriptors: note process_* metrics are generic - Go memory usage: fix description (sys_bytes is runtime memory, not host) - Go goroutine spike: use deriv() instead of rate() on gauge - Go GC CPU fraction: note deprecation since Go 1.20 - Go GC duration: clarify quantile="1" is max, not p99 - Python uncollectable: use increase() on counter instead of raw threshold - Add threshold comments for workload-dependent defaults
This commit is contained in:
Samuel Berthe
parent
127b19714e
commit
0e21e48a59
1 changed files with 34 additions and 4 deletions
|
|
@ -2144,9 +2144,12 @@ groups:
|
|||
for: 2m
|
||||
- name: JVM non-heap memory filling up
|
||||
description: JVM non-heap memory (metaspace/code cache) is filling up (> 80%)
|
||||
query: '(sum by (instance)(jvm_memory_used_bytes{area="nonheap"}) / sum by (instance)(jvm_memory_max_bytes{area="nonheap"})) * 100 > 80'
|
||||
query: '(sum by (instance)(jvm_memory_used_bytes{area="nonheap"}) / (sum by (instance)(jvm_memory_max_bytes{area="nonheap"}) > 0)) * 100 > 80'
|
||||
severity: warning
|
||||
for: 2m
|
||||
comments: |
|
||||
Many JVM configurations leave metaspace unbounded, in which case jvm_memory_max_bytes{area="nonheap"} is -1 and this alert will not fire.
|
||||
The query filters out max_bytes <= 0 to avoid false negatives.
|
||||
- name: JVM GC time too high
|
||||
description: JVM is spending too much time in garbage collection (> 5% of wall clock time)
|
||||
query: 'sum by (instance)(rate(jvm_gc_collection_seconds_sum[5m])) > 0.05'
|
||||
|
|
@ -2172,6 +2175,9 @@ groups:
|
|||
query: 'rate(jvm_gc_collection_seconds_count{gc=~".*old.*|.*major.*"}[5m]) > 0.3'
|
||||
severity: warning
|
||||
for: 5m
|
||||
comments: |
|
||||
This regex matches CMS, G1, and Parallel collector names. It will not match ZGC or Shenandoah cycle names.
|
||||
Adjust the gc label filter if you use a different collector.
|
||||
- name: JVM direct buffer pool filling up
|
||||
description: JVM direct buffer pool is filling up (> 90%)
|
||||
query: '(jvm_buffer_pool_used_bytes / jvm_buffer_pool_capacity_bytes) * 100 > 90'
|
||||
|
|
@ -2187,6 +2193,9 @@ groups:
|
|||
query: '(process_open_fds / process_max_fds) * 100 > 90'
|
||||
severity: warning
|
||||
for: 5m
|
||||
comments: |
|
||||
process_open_fds and process_max_fds are generic metrics from the Prometheus client library, not JVM-specific.
|
||||
This alert will also fire for Go, Python, or any process exposing these metrics.
|
||||
- name: JVM class loading anomaly
|
||||
description: Rapid class loading detected, potential classloader leak
|
||||
query: 'rate(jvm_classes_loaded_total[5m]) > 100'
|
||||
|
|
@ -2209,34 +2218,49 @@ groups:
|
|||
query: 'go_goroutines > 1000'
|
||||
severity: warning
|
||||
for: 5m
|
||||
comments: |
|
||||
Threshold is a rough default. High-concurrency servers may legitimately run thousands of goroutines. Adjust to match your baseline.
|
||||
- name: Go GC duration high
|
||||
description: Go GC pause duration is too high (max > 1s)
|
||||
query: 'go_gc_duration_seconds{quantile="1"} > 1'
|
||||
severity: warning
|
||||
for: 5m
|
||||
comments: |
|
||||
quantile="1" is the maximum observed GC pause in the current summary window, not p99.
|
||||
A single outlier pause can push this above 1s. The for: 5m ensures the max stays elevated.
|
||||
- name: Go memory usage high
|
||||
description: Go heap memory usage is high (> 90% of system memory)
|
||||
description: Go heap allocation is using most of the runtime's reserved memory (> 90%), indicating the process may need more memory or has a leak
|
||||
query: '(go_memstats_heap_alloc_bytes / go_memstats_sys_bytes) * 100 > 90'
|
||||
severity: warning
|
||||
for: 5m
|
||||
comments: |
|
||||
go_memstats_sys_bytes is the total memory obtained from the OS by the Go runtime, not total host memory.
|
||||
This ratio measures Go-internal memory utilization, not system-level memory pressure.
|
||||
- name: Go thread count high
|
||||
description: Go OS thread count is high (> 50), potential blocking syscall or CGo leak
|
||||
query: 'go_threads > 50'
|
||||
severity: warning
|
||||
for: 5m
|
||||
comments: |
|
||||
Threshold is workload-dependent. Applications with heavy CGo or blocking I/O may legitimately use more OS threads. Adjust to match your baseline.
|
||||
- name: Go heap objects count high
|
||||
description: Go heap has too many live objects (> 10M), high GC pressure
|
||||
query: 'go_memstats_heap_objects > 10000000'
|
||||
severity: warning
|
||||
for: 5m
|
||||
comments: |
|
||||
Threshold is a rough default. Adjust based on your application's normal object count.
|
||||
- name: Go GC CPU fraction high
|
||||
description: Go GC is consuming too much CPU (> 5%)
|
||||
query: 'go_memstats_gc_cpu_fraction > 0.05'
|
||||
severity: warning
|
||||
for: 5m
|
||||
comments: |
|
||||
go_memstats_gc_cpu_fraction is deprecated since Go 1.20 and may return 0 in newer versions.
|
||||
Consider using runtime/metrics-based alternatives if running Go >= 1.20.
|
||||
- name: Go goroutine spike
|
||||
description: Go goroutine count is growing rapidly
|
||||
query: 'rate(go_goroutines[5m]) > 100'
|
||||
query: 'deriv(go_goroutines[5m]) > 100'
|
||||
severity: warning
|
||||
for: 5m
|
||||
- name: Go heap fragmentation
|
||||
|
|
@ -2266,6 +2290,8 @@ groups:
|
|||
query: 'ruby_heap_live_slots > 500000'
|
||||
severity: warning
|
||||
for: 5m
|
||||
comments: |
|
||||
Threshold is a rough default. Adjust based on your application's normal heap size.
|
||||
- name: Ruby heap free slots high
|
||||
description: Ruby heap has too many free slots (> 500k), memory fragmentation after large allocations
|
||||
query: 'ruby_heap_free_slots > 500000'
|
||||
|
|
@ -2295,7 +2321,7 @@ groups:
|
|||
rules:
|
||||
- name: Python GC objects uncollectable
|
||||
description: Python has uncollectable objects, potential memory leak via reference cycles
|
||||
query: 'python_gc_objects_uncollectable_total > 0'
|
||||
query: 'increase(python_gc_objects_uncollectable_total[5m]) > 0'
|
||||
severity: warning
|
||||
for: 5m
|
||||
- name: Python GC collections high
|
||||
|
|
@ -2308,6 +2334,8 @@ groups:
|
|||
query: '(process_open_fds / process_max_fds) * 100 > 90'
|
||||
severity: warning
|
||||
for: 5m
|
||||
comments: |
|
||||
process_open_fds and process_max_fds are generic metrics from the Prometheus client library, not Python-specific.
|
||||
- name: Python GC generation 2 collections high
|
||||
description: Python full GC (generation 2) is running too frequently, indicating memory pressure
|
||||
query: 'rate(python_gc_collections_total{generation="2"}[5m]) > 1'
|
||||
|
|
@ -2318,6 +2346,8 @@ groups:
|
|||
query: 'process_virtual_memory_bytes > 4e9'
|
||||
severity: warning
|
||||
for: 5m
|
||||
comments: |
|
||||
Threshold is a rough default. Adjust based on your application's expected memory footprint.
|
||||
|
||||
- name: Sidekiq
|
||||
exporters:
|
||||
|
|
|
|||
Loading…
Reference in a new issue