# Limits and defaults

Hard limits, soft limits and default sizes: queue lengths, timeouts, retention.

Last reviewed 2026-10-10

The numbers you can ignore until scale finds them. The full set is in [Configuration reference](https://n9e.github.io/docs/reference/configuration.md).

## HTTP[​](#http "Direct link to HTTP")

| Item               | Default | Key                       |
| ------------------ | ------- | ------------------------- |
| Request body limit | 64 MiB  | `[HTTP] MaxContentLength` |
| Read timeout       | 20s     | `[HTTP] ReadTimeout`      |
| Write timeout      | 40s     | `[HTTP] WriteTimeout`     |
| Idle timeout       | 120s    | `[HTTP] IdleTimeout`      |
| Graceful shutdown  | 30s     | `[HTTP] ShutdownTimeout`  |

Bulk-importing alert rules or a large dashboard JSON is what usually meets the body limit first.

## Metadata database[​](#metadata-database "Direct link to Metadata database")

| Item                    | Default | Key                 |
| ----------------------- | ------- | ------------------- |
| Max open connections    | 150     | `[DB] MaxOpenConns` |
| Max idle connections    | 50      | `[DB] MaxIdleConns` |
| Max connection lifetime | 7200s   | `[DB] MaxLifetime`  |

Multiply by the number of instances and stay under the database's own `max_connections`.

## Embedded TSDB[​](#embedded-tsdb "Direct link to Embedded TSDB")

| Item          | Default | Key                                |
| ------------- | ------- | ---------------------------------- |
| Retention     | 15d     | `[EmbeddedTSDB] RetentionDuration` |
| Disk cap      | 10 GiB  | `[EmbeddedTSDB] MaxBytes`          |
| Query timeout | 1m      | `[EmbeddedTSDB] QueryTimeout`      |

Past `MaxBytes`, **the oldest blocks are deleted first**.

The real limit is not any of these numbers: it is that the data lives on **one Center process's local disk**, so several replicas each hold a fragment. Useful to roughly 100k active series; past that, use an external store. See [Embedded TSDB single-Center limits](https://n9e.github.io/docs/operations/embedded-tsdb-limits.md).

## Ingest and forwarding[​](#ingest-and-forwarding "Direct link to Ingest and forwarding")

| Item                    | Default | Key                              |
| ----------------------- | ------- | -------------------------------- |
| Forward queue length    | 1000000 | `[Pushgw] QueueMaxSize`          |
| Forward request timeout | 10000ms | `[[Pushgw.Writers]] Timeout`     |
| Dial timeout            | 3000ms  | `[[Pushgw.Writers]] DialTimeout` |

A full queue drops data. A queue length that keeps climbing toward the cap means the downstream TSDB cannot keep up; watching `n9e_alert_alert_queue_size` and friends catches it earlier — see [Built-in metrics](https://n9e.github.io/docs/reference/built-in-metrics.md).

## Rule evaluation[​](#rule-evaluation "Direct link to Rule evaluation")

| Item                | Default       | Where     |
| ------------------- | ------------- | --------- |
| Execution frequency | `@every 60s`  | Rule form |
| For duration        | 60s           | Rule form |
| Recover duration    | 0 (immediate) | Rule form |
| Repeat interval     | 60 minutes    | Rule form |
| Max send times      | 0 (unlimited) | Rule form |

## Data retention[​](#data-retention "Direct link to Data retention")

| Data                       | Kept for           | Key                                |
| -------------------------- | ------------------ | ---------------------------------- |
| Notification records       | 7 days             | `[Center] CleanNotifyRecordDay`    |
| Workflow execution records | 7 days             | built into the cleanup job         |
| Embedded TSDB samples      | 15 days            | `[EmbeddedTSDB] RetentionDuration` |
| Historical alert events    | forever by default | `[Center] CleanAlertHisEventDay`   |

**Historical alert events are kept forever unless you say otherwise.** The knob exists but ships commented out, and `<= 0` means keep forever. Set `CleanAlertHisEventDay` to a number of days and a daily 02:00 job deletes older events in batches — otherwise plan for that table's growth.

## Related[​](#related "Direct link to Related")

* [Configuration reference](https://n9e.github.io/docs/reference/configuration.md)
* [Capacity planning](https://n9e.github.io/docs/operations/capacity.md)
* [Why is Nightingale slow or backlogged?](https://n9e.github.io/docs/troubleshooting/performance.md)