> ## Documentation Index
> Fetch the complete documentation index at: https://doc.lucidworks.com/llms.txt
> Use this file to discover all available pages before exploring further.

# Fusion 5.18 patch releases

This page lists all patch releases for Fusion 5.18, starting with the most recent. For the 5.18.0 minor release, see [Fusion 5.18.0](https://doc.lucidworks.com/docs/5/fusion/release-notes/5.18/5.18.0-release-notes.md).

To get notified when new Fusion 5.18 patch releases are published, select the **RSS** icon at the top of this page and add the feed to your RSS reader.

**5.18.1**

November 16, 2026

Released on November 16, 2026, Fusion 5.18.1 is a [patch release](https://doc.lucidworks.com/docs/policies/lifecycle-policies/lw-semantic-version-support-lifecycle.md#patch-release-support-policy) that delivers security updates for API Gateway authentication and Solr local parameters, and bug fixes for API Gateway latency, connector recrawls, stray content deletion, Spark jobs in TLS-enabled clusters, and job scheduling.

## Security updates

### The API Gateway accepted blank usernames, realm names, and JWT subjects

In previous versions of Fusion, the API Gateway accepted *JSON Web Tokens* (*JWTs*) that had a blank or missing subject claim, and the API Gateway accepted requests to create users and security realms that had blank names.

Fusion 5.18.1 now rejects JWTs whose subject claim is blank or missing, and rejects requests to create a user or security realm that has a blank name. At startup, Fusion 5.18.1 logs and skips blank entries in the initial roles, realms, and users configuration. When Fusion 5.18.1 creates the default `webapps-jwt-realm` security realm on a new cluster, it generates a unique signing key for the realm.

> [accordion] Supply your own JWT signing key
>
> To supply your own signing key for the `webapps-jwt-realm` security realm on a new cluster, do the following:
>
> 1. Create a Kubernetes Secret that contains a signing key of at least 32 characters.
> 2. In the `api-gateway` Helm values, set the `initialJwtSigningKey.existingSecret` value to the name of the Secret.
> 3. Set the `initialJwtSigningKey.secretKey` value to the Secret entry that contains the signing key.
>
> Fusion reads this key only when it creates the realm, so clusters upgraded to Fusion 5.18.1 keep their existing signing key. To replace the key on an upgraded cluster, see [Security best practices](https://doc.lucidworks.com/docs/5/fusion/reference/config-ref/configure-fusion-for-jwt-5x.md#security-best-practices) in "Configure Fusion for JWT."

### Query requests could pass unsafe Solr local parameters to Solr

In previous versions of Fusion, query pipelines passed Solr local parameters in caller-supplied request parameters to Solr without checking them. A caller could use a cross-collection join or a `zkHost` or `solrUrl` local parameter to make Solr open connections to external hosts, and the failed connections could accumulate and slow queries. In restrictive mode, Fusion also didn't reject caller-supplied local parameters in some request parameters, such as `facet.field`, `bq`, `fl`, and `sort`.

Fusion 5.18.1 now rejects a query request with an HTTP 400 response when a request parameter or the JSON request body contains a `crossCollection` join, a `zkHost` or `solrUrl` local parameter, or the `xmlparser` query parser. Other local parameters, such as `tag`, `ex`, and `key`, and joins within the same cluster still work. Local parameters that query pipeline stages, query profiles, and rules add aren't checked. In restrictive mode, Fusion 5.18.1 also rejects caller-supplied local parameters in `facet.field`, `facet.query`, `bq`, `boost`, `bf`, `fl`, `sort`, and `json.facet`, and validates requests to the `/executions` and `/preview` endpoints.

> [!NOTE]
>
> This check is on by default. To turn it off, set the `localParamsGuard.enabled` value to `false` in the `query-pipeline` Helm values.

## Bug fixes

### The API Gateway added latency to queries under load

In Fusion 5.17.2 and 5.18.0, the API Gateway added up to several seconds of latency to some requests under sustained query load, even when the execution times of query pipelines and Solr were normal.

Fusion 5.18.1 removes thread contention in API Gateway request routing and token handling, so API Gateway response times stay in line with the execution times of query pipelines under load.

### The first recrawl after upgrading to Fusion 5.18.0 reindexed all documents

In Fusion 5.18.0, the first recrawl after upgrading from Fusion 5.9 through 5.17 treated every document in each affected datasource as changed and sent it through indexing again, because those earlier versions don't store document signatures.

Fusion 5.18.1 turns off document signature evaluation, so upgrading from Fusion 5.9 through 5.17 no longer causes a full reindex on the first recrawl. Documents indexed by Fusion 5.18.1 don't receive the `_lw_signature_s` field, and documents indexed earlier keep the field until they're reindexed.

### Web V2, file system, and SMB connectors re-fetched every item on every recrawl

In Fusion 5.18.0, Web V2, file system, and SMB connectors didn't receive the stored state that they use to skip unchanged items, such as `ETag` and `Last-Modified` header values for Web V2 or modification times for files. As a result, every recrawl re-fetched every page or re-read every file instead of only changed items.

Fusion 5.18.1 replays stored state to connectors on recrawl, so they fetch only new and changed items. The first recrawl of a datasource that was last crawled on Fusion 5.18.0 re-fetches every item once, because Fusion 5.18.0 didn't store the state that replay needs.

> [accordion] Return to the Fusion 5.18.0 recrawl behavior
>
> To return to the Fusion 5.18.0 recrawl behavior, globally or for specific datasources, set these JVM system properties in the `javaToolOptions` Helm value for the `connectors-backend` service:
>
> ```java wrap theme={"dark"}
> -Dcom.lucidworks.connectors.recrawl.mode=replay|rediscover
> -Dcom.lucidworks.connectors.recrawl.mode.rediscoverDatasourceIds=<id1>,<id2>
> -Dcom.lucidworks.connectors.recrawl.mode.replayDatasourceIds=<id3>,<id4>
> ```
>
> To re-fetch every item on recrawl, set the `com.lucidworks.connectors.recrawl.mode` property to `rediscover`. To fetch only new and changed items, leave the property at its default value, `replay`. To override the global value for a datasource, add the datasource ID to the `rediscoverDatasourceIds` or `replayDatasourceIds` property. To apply your changes, restart the service.

### Crawls started over after a failed or stopped job

In Fusion 5.18.0, when a crawl job failed or was stopped, the next run discarded the job's checkpoints and crawled the entire source again.

Fusion 5.18.1 saves a snapshot of a job's checkpoints when the job completes. When the previous job didn't complete, the next run resumes from that snapshot instead of crawling the entire source again. Connectors that don't use checkpoints aren't affected.

### Stray content deletion removed valid content and left orphaned documents

In Fusion 5.18.0, stray content deletion could remove the CrawlDB entries for parsed child documents, such as rows of a CSV file, when their parent file was unchanged. Those child documents were then orphaned in the content collection, and stray content deletion could also delete indexed content that was still valid.

Fusion 5.18.1 keeps the CrawlDB entries for parsed child documents of unchanged parent files, so stray content deletion deletes only content that was removed from the source.

### Stray content deletion silently stopped on large batches

In Fusion 5.18.0, when a crawl had many stray items, the request that counts the documents to delete could exceed the server's HTTP header size limit. The cleanup then stopped without failing the job, and stray content stayed in the content collection.

Fusion 5.18.1 sends this request in the request body instead of the URL, so stray content deletion completes for large batches.

### Upgrading turned on stray content deletion for AWS S3 Pro and JDBC V2 datasources that had it turned off

In Fusion 5.18.0, upgraded or imported AWS S3 Pro and JDBC V2 datasources deleted stray content even when they had turned it off before Fusion 5.18, with the `enableStrayContentDeletion` property for AWS S3 Pro, or with the `strayContentDeletionEnabled` property or a delta query for JDBC V2. Fusion 5.18.0 reads only the `coreProperties.strayDeletionSettings.enableStrayDeletion` property, which defaults to `true`.

Fusion 5.18.1 copies the earlier setting to the `enableStrayDeletion` property when the `connectors-backend` service starts and whenever an AWS S3 Pro or JDBC V2 datasource is created or updated. If the `enableStrayDeletion` property is already set, Fusion 5.18.1 doesn't change it, and logs a warning when it's `true` but the earlier setting turned stray content deletion off.

### Spark jobs failed in TLS-enabled clusters

In Fusion 5.18.0, Spark jobs in TLS-enabled Fusion clusters could fail or time out. Spark driver pods didn't exit after their jobs finished, and Spark executors sent requests to the proxy service over HTTP instead of HTTPS, so jobs that index documents, such as script jobs and Parallel Bulk Loader jobs, failed.

Fusion 5.18.1 now shuts down Spark drivers when their jobs finish and sends executor requests to the proxy address that the job launcher configures, so Spark jobs complete in TLS-enabled clusters.

### Completed jobs and their chained jobs ran again after job services restarted

In previous versions of Fusion, the `job-config` service started a job's chained jobs every time it received a completion event for the job run. After the `job-config` and `job-launcher` services restarted, duplicate completion events could start chained jobs again and start completed jobs again.

Fusion 5.18.1 starts chained jobs only once for each job run, including after the `job-config` service restarts.

### Scheduled jobs stopped running when Kafka was unavailable at startup

In previous versions of Fusion, when Kafka was unavailable while the `job-config` service started, the service could keep the scheduler leadership without starting the scheduler. Scheduled jobs then stopped running until the `job-config` pod was restarted.

Fusion 5.18.1 now reports the `job-config` service as unhealthy when the scheduler doesn't start, so Kubernetes restarts the pod and the scheduler starts on a healthy pod.

## Known issues

### LWAI Vectorize Query and LWAI Prediction stages fail with asynchronous execution

In Fusion 5.18.1, the LWAI Vectorize Query and LWAI Prediction query stages don't work when **Enable Async Execution** is selected in the stage's **Asynchronous Execution Config**.

**Impact:** An LWAI Vectorize Query stage doesn't produce a vector, so the query doesn't include a vector search. An LWAI Prediction stage fails with an error from the Lucidworks AI Gateway.

**Workaround:** Clear the **Enable Async Execution** checkbox in the LWAI Vectorize Query and LWAI Prediction stages.


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.