|
|
|
|
@@ -188,7 +188,7 @@ If you see unexpected or unreliable query results from VictoriaMetrics, then try
|
|
|
|
|
|
|
|
|
|
These are the most common reasons for slow data ingestion in VictoriaMetrics:
|
|
|
|
|
|
|
|
|
|
1. [Memory shortage](https://docs.victoriametrics.com/victoriametrics/troubleshooting/#memory-shortage) for the given amounts of [active time series](https://docs.victoriametrics.com/victoriametrics/faq/#what-is-an-active-time-series).
|
|
|
|
|
1. Memory shortage for the given amounts of [active time series](https://docs.victoriametrics.com/victoriametrics/faq/#what-is-an-active-time-series).
|
|
|
|
|
|
|
|
|
|
VictoriaMetrics (or `vmstorage` in the cluster version of VictoriaMetrics) maintains an in-memory cache `storage/tsid`
|
|
|
|
|
for a quick search for internal series IDs for each incoming metric. VictoriaMetrics automatically determines the maximum
|
|
|
|
|
@@ -352,138 +352,35 @@ These are the solutions that exist for improving the performance of slow queries
|
|
|
|
|
See also [this article](https://valyala.medium.com/how-to-optimize-promql-and-metricsql-queries-85a1b75bf986),
|
|
|
|
|
which explains how to identify and optimize slow queries.
|
|
|
|
|
|
|
|
|
|
## Memory shortage
|
|
|
|
|
## Out of memory errors
|
|
|
|
|
|
|
|
|
|
High memory utilization alone does not indicate a shortage.
|
|
|
|
|
A VictoriaMetrics component can operate normally under high memory utilization,
|
|
|
|
|
but it is recommended to keep [at least 50% of free memory for stability](https://docs.victoriametrics.com/victoriametrics/single-server-victoriametrics/#capacity-planning).
|
|
|
|
|
A shortage means there is not enough memory for the workload.
|
|
|
|
|
It is different from high utilization and from memory pressure (the kernel's reclaim activity, shown by [PSI](https://docs.kernel.org/accounting/psi.html)).
|
|
|
|
|
Use the [signals](https://docs.victoriametrics.com/victoriametrics/troubleshooting/#memory-signals) and [patterns](https://docs.victoriametrics.com/victoriametrics/troubleshooting/#memory-shortage-patterns) below to tell them apart,
|
|
|
|
|
and [how to fix](https://docs.victoriametrics.com/victoriametrics/troubleshooting/#how-to-fix-memory-issues) to resolve it.
|
|
|
|
|
The following are the most common sources of out-of-memory (aka OOM) crashes in VictoriaMetrics:
|
|
|
|
|
|
|
|
|
|
VictoriaMetrics components detect the available memory at startup as the smaller of the host RAM and the cgroup memory limit,
|
|
|
|
|
and expose it as `vm_available_memory_bytes`. The actual usage is `process_resident_memory_bytes`, which has two main parts:
|
|
|
|
|
1. Improper command-line flag values. Inspect command-line flags passed to VictoriaMetrics components.
|
|
|
|
|
If you don't clearly understand the purpose or the effect of some flags, remove them
|
|
|
|
|
from the list of flags passed to VictoriaMetrics components. Improper command-line flag values
|
|
|
|
|
may lead to increased memory and CPU usage. Increased memory usage increases the risk of OOM crashes.
|
|
|
|
|
VictoriaMetrics is optimized to run with default flag values (e.g., when they aren't explicitly set).
|
|
|
|
|
|
|
|
|
|
1. Go (anonymous) memory - `process_resident_memory_anon_bytes`. It includes:
|
|
|
|
|
For example, it isn't recommended to change cache sizes in VictoriaMetrics, as this frequently leads to OOM exceptions.
|
|
|
|
|
[These docs](https://docs.victoriametrics.com/victoriametrics/single-server-victoriametrics/#cache-tuning) refer to command-line flags that aren't
|
|
|
|
|
recommended to tune. If you see that VictoriaMetrics needs to increase some cache sizes for the current workload,
|
|
|
|
|
then it is better to migrate to a host with more memory instead of trying to tune cache sizes manually.
|
|
|
|
|
|
|
|
|
|
- `-memory.allowedPercent` (default 60%, or `-memory.allowedBytes`) sets a memory budget whose use differs per component:
|
|
|
|
|
- `vmstorage`: in-process caches (for example `storage/tsid` and `indexdb/file`) and in-memory data parts.
|
|
|
|
|
- `vmselect`: the rollup result cache and the per-query rollup memory.
|
|
|
|
|
- `vminsert`: in-memory row buffers held per `vmstorage` node before flushing to `vmstorage`.
|
|
|
|
|
- `vmagent`: in-memory blocks held before they are written to the persistent queue.
|
|
|
|
|
- `vmauth`: not sized by it, but bounded by `-maxConcurrentRequests` and `-requestBufferSize` instead.
|
|
|
|
|
- The Go heap, goroutine stacks and runtime overhead used for ingestion and queries.
|
|
|
|
|
1. Unexpected heavy queries. The query is considered heavy if it needs to select and process millions of unique time series.
|
|
|
|
|
Such a query may cause an OOM exception, as VictoriaMetrics needs to keep some per-series data in memory.
|
|
|
|
|
VictoriaMetrics provides [various settings](https://docs.victoriametrics.com/victoriametrics/single-server-victoriametrics/#resource-usage-limits)
|
|
|
|
|
that can help limit resource usage.
|
|
|
|
|
For more context, see [How to optimize PromQL and MetricsQL queries](https://valyala.medium.com/how-to-optimize-promql-and-metricsql-queries-85a1b75bf986).
|
|
|
|
|
VictoriaMetrics also provides [query tracer](https://docs.victoriametrics.com/victoriametrics/single-server-victoriametrics/#query-tracing)
|
|
|
|
|
to help identify the source of heavy queries. Slow queries can be logged with additional details via [Query execution stats](https://docs.victoriametrics.com/victoriametrics/query-stats/).
|
|
|
|
|
|
|
|
|
|
1. OS page cache for the on-disk [data](https://docs.victoriametrics.com/victoriametrics/#storage)
|
|
|
|
|
and [indexdb](https://docs.victoriametrics.com/victoriametrics/#indexdb). The OS caches recently
|
|
|
|
|
read parts of these files in free RAM and reclaims them under memory pressure.
|
|
|
|
|
`process_resident_memory_file_bytes` shows how much of them is currently resident for the process.
|
|
|
|
|
|
|
|
|
|
Before tuning and troubleshooting memory issues,
|
|
|
|
|
see [Best practices](https://docs.victoriametrics.com/victoriametrics/bestpractices/#memory)
|
|
|
|
|
for memory configuration guidance. Be sure that `GOMEMLIMIT` is not set,
|
|
|
|
|
and that the [VPA](https://github.com/kubernetes/autoscaler/tree/master/vertical-pod-autoscaler) is not used for `vmstorage` pods.
|
|
|
|
|
|
|
|
|
|
### Memory signals
|
|
|
|
|
|
|
|
|
|
These metrics describe how a component uses memory. See [how to monitor VictoriaMetrics](https://docs.victoriametrics.com/victoriametrics/#monitoring)
|
|
|
|
|
to set up scraping and the Grafana dashboards that show them.
|
|
|
|
|
None of them means a shortage on its own; read them together in [Memory shortage patterns](https://docs.victoriametrics.com/victoriametrics/troubleshooting/#memory-shortage-patterns).
|
|
|
|
|
|
|
|
|
|
- `process_resident_memory_anon_bytes` / `vm_available_memory_bytes` - anonymous memory (caches plus Go heap)
|
|
|
|
|
as a share of the available memory. This memory can't be reclaimed back by OS.
|
|
|
|
|
|
|
|
|
|
- `process_resident_memory_file_bytes` - the OS page cache for the component's on-disk data,
|
|
|
|
|
currently resident for the process. Reclaimable by the OS.
|
|
|
|
|
|
|
|
|
|
- `process_pressure_memory_waiting_seconds_total`, `process_pressure_memory_stalled_seconds_total` -
|
|
|
|
|
[PSI](https://docs.kernel.org/accounting/psi.html): time tasks were stalled waiting for memory reclaim.
|
|
|
|
|
Populated only on Linux hosts with PSI support.
|
|
|
|
|
|
|
|
|
|
- `vm_cache_size_bytes / vm_cache_size_max_bytes` (per `type`, e.g., `storage/tsid`) -
|
|
|
|
|
how full each in-process cache is.
|
|
|
|
|
|
|
|
|
|
- `vm_slow_row_inserts_total` / `vm_rows_added_to_storage_total` -
|
|
|
|
|
share of ingested rows that missed the `storage/tsid` cache ([slow inserts](https://docs.victoriametrics.com/victoriametrics/troubleshooting/#slow-data-ingestion)).
|
|
|
|
|
|
|
|
|
|
- `increase(vm_new_timeseries_created_total[24h]) / vm_cache_entries{type="storage/hour_metric_ids"}` -
|
|
|
|
|
the [churn rate](https://docs.victoriametrics.com/victoriametrics/faq/#what-is-high-churn-rate):
|
|
|
|
|
new series created over a day relative to the
|
|
|
|
|
[active time series](https://docs.victoriametrics.com/victoriametrics/faq/#what-is-an-active-time-series).
|
|
|
|
|
|
|
|
|
|
- `process_major_pagefaults_total` - rate of pages read from disk (page-cache misses, refaults, or swap-in).
|
|
|
|
|
|
|
|
|
|
- `go_memstats_heap_inuse_bytes` and the `CPU spent on GC` panel - the Go heap working set and the CPU cost of garbage collection.
|
|
|
|
|
|
|
|
|
|
#### Out of memory errors
|
|
|
|
|
|
|
|
|
|
An out-of-memory (OOM) kill is the strongest sign of a memory shortage, but the process cannot report it:
|
|
|
|
|
the process is already dead. Detect the kill from outside the process:
|
|
|
|
|
|
|
|
|
|
- Kubernetes: a container restart with reason `OOMKilled` in the pod events (`kubectl describe pod`).
|
|
|
|
|
- Linux hosts: the kernel OOM killer log in `dmesg` or `journalctl`, with the `oom_score` for the killed process.
|
|
|
|
|
- Container runtime logs record the same kill from the runtime side.
|
|
|
|
|
|
|
|
|
|
To prevent recurrence, resolve the underlying shortage (see [How to fix memory issues](https://docs.victoriametrics.com/victoriametrics/troubleshooting/#how-to-fix-memory-issues)).
|
|
|
|
|
|
|
|
|
|
### Memory shortage patterns
|
|
|
|
|
|
|
|
|
|
There are three patterns of memory shortage:
|
|
|
|
|
|
|
|
|
|
1. **The cache cannot hold the active series (cache-bound shortage).** The `storage/tsid` cache is full:
|
|
|
|
|
`vm_cache_size_bytes{type="storage/tsid"}` is close to `vm_cache_size_max_bytes{type="storage/tsid"}`, and slow inserts stay high.
|
|
|
|
|
Most slow inserts come from `storage/tsid` cache misses, on new series or on already-known active series.
|
|
|
|
|
If they stay above 5% of ingested rows during a stable window without restarts or rerouting,
|
|
|
|
|
and are not explained by `rate(vm_new_timeseries_created_total)`,
|
|
|
|
|
it points to misses on active series that no longer fit the cache.
|
|
|
|
|
See the detailed explanation in the [Slow data ingestion](https://docs.victoriametrics.com/victoriametrics/troubleshooting/#slow-data-ingestion) section.
|
|
|
|
|
|
|
|
|
|
1. **The Go heap exceeds its budget (heap-bound shortage).** `go_memstats_heap_inuse_bytes` climbs well above its stable baseline.
|
|
|
|
|
Check it together with the non-cache part of anonymous memory:
|
|
|
|
|
`process_resident_memory_anon_bytes` minus the component's `vm_cache_size_bytes`.
|
|
|
|
|
There is no fixed normal value, so compare against the component's own stable history.
|
|
|
|
|
A `process_resident_memory_anon_bytes / vm_available_memory_bytes` ratio that keeps rising leaves little headroom
|
|
|
|
|
and may lead to an [OOM kill](https://docs.victoriametrics.com/victoriametrics/troubleshooting/#out-of-memory-errors).
|
|
|
|
|
A single heavy query can spike the heap on its own: if it has to select and process millions of unique time series,
|
|
|
|
|
VictoriaMetrics keeps some per-series data in memory while the query runs. Find it with the
|
|
|
|
|
[query tracer](https://docs.victoriametrics.com/victoriametrics/single-server-victoriametrics/#query-tracing),
|
|
|
|
|
log slow queries with [query execution stats](https://docs.victoriametrics.com/victoriametrics/query-stats/),
|
|
|
|
|
bound it with [resource usage limits](https://docs.victoriametrics.com/victoriametrics/single-server-victoriametrics/#resource-usage-limits),
|
|
|
|
|
and see [how to optimize PromQL and MetricsQL queries](https://valyala.medium.com/how-to-optimize-promql-and-metricsql-queries-85a1b75bf986).
|
|
|
|
|
If heap growth correlates with query or ingestion load, it is workload-driven.
|
|
|
|
|
If the heap grows regardless of load, suspect a memory leak, collect a heap profile, and [file a bug report](https://github.com/VictoriaMetrics/VictoriaMetrics/issues/new).
|
|
|
|
|
|
|
|
|
|
1. **The OS page cache is too small (I/O-bound shortage).** The main signal is a consistently high
|
|
|
|
|
`process_major_pagefaults_total` rate: the component's data and `indexdb` no longer fit the page cache,
|
|
|
|
|
so it reads them from disk and query latency grows. `process_resident_memory_file_bytes` drops as the OS
|
|
|
|
|
evicts these file pages. Swap causes the same symptoms, so keep it disabled on `vmstorage` and
|
|
|
|
|
single-node hosts (see [Swap](https://docs.victoriametrics.com/victoriametrics/bestpractices/#swap)).
|
|
|
|
|
Add memory according to
|
|
|
|
|
[capacity planning](https://docs.victoriametrics.com/victoriametrics/single-server-victoriametrics/#capacity-planning).
|
|
|
|
|
|
|
|
|
|
PSI is an additional confirmation signal: a rising `process_pressure_memory_*` value indicates that the kernel is reclaiming memory for the cgroup.
|
|
|
|
|
Where PSI is unavailable, rely on the per-pattern signals above.
|
|
|
|
|
|
|
|
|
|
### How to fix memory issues
|
|
|
|
|
|
|
|
|
|
After distinguishing the shortage from normal high memory utilization, and if it persists,
|
|
|
|
|
you can use the approaches below to resolve it:
|
|
|
|
|
|
|
|
|
|
- Reduce the number of [active time series](https://docs.victoriametrics.com/victoriametrics/faq/#what-is-an-active-time-series)
|
|
|
|
|
or the [churn rate](https://docs.victoriametrics.com/victoriametrics/faq/#what-is-high-churn-rate) -
|
|
|
|
|
see [Slow data ingestion](https://docs.victoriametrics.com/victoriametrics/troubleshooting/#slow-data-ingestion).
|
|
|
|
|
- Add more memory by scaling vertically or horizontally - see capacity planning for
|
|
|
|
|
[single-node VictoriaMetrics](https://docs.victoriametrics.com/victoriametrics/single-server-victoriametrics/#capacity-planning)
|
|
|
|
|
and the [cluster version](https://docs.victoriametrics.com/victoriametrics/cluster-victoriametrics/#capacity-planning).
|
|
|
|
|
Spare memory also absorbs workload spikes that would otherwise OOM a component running near its limit.
|
|
|
|
|
- Remove command-line flags whose impact you do not clearly understand. Improper flags can raise
|
|
|
|
|
memory usage and lead to OOM crashes. In particular, do not change
|
|
|
|
|
[cache sizes](https://docs.victoriametrics.com/victoriametrics/single-server-victoriametrics/#cache-tuning);
|
|
|
|
|
add more memory instead.
|
|
|
|
|
- Investigate Go heap growth or a suspected memory leak - collect a memory profile using the profiling guide for
|
|
|
|
|
[single-node VictoriaMetrics](https://docs.victoriametrics.com/victoriametrics/single-server-victoriametrics/#profiling)
|
|
|
|
|
or [cluster components](https://docs.victoriametrics.com/victoriametrics/cluster-victoriametrics/#profiling).
|
|
|
|
|
1. Lack of free memory for processing workload spikes. If VictoriaMetrics components use almost all the available memory
|
|
|
|
|
under the current workload, then it is recommended to migrate to a host with larger amounts of memory.
|
|
|
|
|
This would protect from possible OOM crashes on workload spikes. It is recommended to have at least 50%
|
|
|
|
|
of free memory to gracefully handle possible workload spikes.
|
|
|
|
|
See [capacity planning for single-node VictoriaMetrics](https://docs.victoriametrics.com/victoriametrics/single-server-victoriametrics/#capacity-planning)
|
|
|
|
|
and [capacity planning for the cluster version of VictoriaMetrics](https://docs.victoriametrics.com/victoriametrics/cluster-victoriametrics/#capacity-planning).
|
|
|
|
|
|
|
|
|
|
## Cluster instability
|
|
|
|
|
|
|
|
|
|
|
