## Documentation Index

Fetch the complete documentation index at: [/docs/llms.txt](/content/docs/llms.txt)

Use this file to discover all available pages before exploring further.

**Applies to:**

- Plan - Plans: Any
- Deployment - Deployments: Self-hosted
- Use case - Interpreting warning and info logs from Brainstore after configuring or running a data retention policy worker

## Summary

**Issue:** After resolving the `aead::Error` decryption failure and enabling time-based retention, Brainstore logs show `Failed to execute static group by`, `Unable to pushdown filter`, and `Deleted index documents with xact_id < ...` messages.  **Cause:** These are query optimization fallbacks and retention confirmation logs, not errors. They appear during normal Brainstore operation.  **Resolution:** No action required. All three log types indicate expected behavior.

---

## Log message reference

### `Failed to execute static group by, falling back to dynamic`

This is a non-critical warning. It fires when a field (e.g., `error`) has inconsistent types across span documents — sometimes a string, sometimes a nested object.  Brainstore attempts a fast columnar aggregation first. When schema inconsistency prevents it, Brainstore retries using a dynamic aggregation path that handles mixed types. Queries return correct results either way. The dynamic path is slightly slower but functionally equivalent.

---

### `Unable to pushdown filter: Comparison not a field and literal`

This is a non-critical query optimization warning. Brainstore’s index layer can only accelerate simple `field = literal` comparisons. Filters involving `coalesce()` expressions generally cannot be pushed down to the index:

```
coalesce(metadata.tool_choice.function.name, '') = 'select_choice'
```

When this occurs, the optimizer evaluates the filter at the query layer instead. Filters still apply correctly — just without index acceleration. These warnings originate from Braintrust’s own UI queries.  **Exception:** `coalesce(field, 'literal1') != 'literal2'` patterns are automatically rewritten to the null-safe not-equal operator (`field <!=> 'literal2'`), which can be pushed down to the index. If you see this warning for a `!=` comparison that uses `coalesce()`, confirm that both literals are the same type.

---

### `Deleted index documents with xact_id < ...`

This is an info-level confirmation that the retention worker ran successfully. `xact_id` is an internal transaction ID used to identify records outside the retention window. Seeing this log means your retention policy is working correctly. No data loss occurs — only documents older than your configured retention threshold are removed.

---

## Fixing the underlying `aead::Error` (if retention is still not running)

If retention has not yet recovered and the original decryption error persists, the most common remaining cause is a stale `bt_data_plane_service_token` encrypted under an old key.

#### Step 1: Verify key consistency

Confirm `FUNCTION_SECRET_KEY` is identical on both the API host and the Brainstore host. If `SERVICE_TOKEN_SECRET_KEY` is unset, the API falls back to `FUNCTION_SECRET_KEY`. Both services must use the same value.

#### Step 2: Restart both services

Restart the API and Brainstore hosts after confirming the keys are correct.

#### Step 3: Refresh the service token

In **Settings** > [**Service tokens**](/content/app/~/configuration/org/service-tokens/index.html), click **Refresh** on `bt_data_plane_service_token`. This re-encrypts the token under the current key. If the token was refreshed before the keys were corrected, it must be refreshed again after.