Support for redis-cluster #10583

New Issue

GiteaMirror · 2025-11-02T09:11:55-06:00

GiteaMirror commented

2025-11-02 09:11:55 -06:00

Originally created by @pat-s on GitHub (Apr 2, 2023).

Feature Description

Using Gitea with redis-cluster seems not possible as of right now.
When attempting so, the following error is thrown by Gitea

gitea 2023/04/01 23:31:36 ...ers/web/auth/auth.go:312:handleSignInFull() [E] [6428bed8] RegenerateSession: regenerate session: CROSSSLOT Keys in request don't hash to the same slot

This is a unique behavior of redis-cluster compared to a single redis installation and must be accounted for on the client side as described in https://dzone.com/articles/resolved-crossslot-keys-error-in-redis-cluster-mod.

When creating keys that could be used by multi-key operations, use hashtags ({…}) to force the keys into the same hash slot. When the key contains a “{…}” pattern, only the substring between the braces, “{” and “}”, is considered to calculate the hash slot.
For example, the keys {user1}:mykey1 and {user1}:mykey2 are hashed to the same hash-slot because only the string inside the braces “{” and “}”, that is, “user1”, is used to compute the hash-slot.

I can't infer if this would be a simple addition or a breaking change but wanted to drop the information about redis-cluster anyhow as I couldn't find a prior discussion about it.

Screenshots

No response

Originally created by @pat-s on GitHub (Apr 2, 2023). ### Feature Description Using Gitea with `redis-cluster` seems not possible as of right now. When attempting so, the following error is thrown by Gitea ``` gitea 2023/04/01 23:31:36 ...ers/web/auth/auth.go:312:handleSignInFull() [E] [6428bed8] RegenerateSession: regenerate session: CROSSSLOT Keys in request don't hash to the same slot ``` This is a unique behavior of `redis-cluster` compared to a single `redis` installation and must be accounted for on the client side as described in https://dzone.com/articles/resolved-crossslot-keys-error-in-redis-cluster-mod. > When creating keys that could be used by multi-key operations, use hashtags ({…}) to force the keys into the same hash slot. When the key contains a “{…}” pattern, only the substring between the braces, “{” and “}”, is considered to calculate the hash slot. For example, the keys {user1}:mykey1 and {user1}:mykey2 are hashed to the same hash-slot because only the string inside the braces “{” and “}”, that is, “user1”, is used to compute the hash-slot. I can't infer if this would be a simple addition or a breaking change but wanted to drop the information about `redis-cluster` anyhow as I couldn't find a prior discussion about it. ### Screenshots _No response_

GiteaMirror added the type/bug label 2025-11-02 09:11:55 -06:00

GiteaMirror closed this issue

2025-11-02 09:11:55 -06:00

GiteaMirror commented

2025-11-02 09:11:56 -06:00

@lunny commented on GitHub (Apr 2, 2023):

Could you paste your configuration and redis cluster information here?

@lunny commented on GitHub (Apr 2, 2023): Could you paste your configuration and redis cluster information here?

GiteaMirror commented

2025-11-02 09:11:56 -06:00

@pat-s commented on GitHub (Apr 2, 2023):

I just used the default config of the bitnami chart. Important: I used the latest v6 redis (chart: release 7.6.4) as the v7 has some compatibility issues with the underlying Debian image and runs into another issue (unrelated to Gitea).

@pat-s commented on GitHub (Apr 2, 2023): I just used the default config of the [bitnami chart](https://github.com/bitnami/charts/tree/main/bitnami/redis-cluster). Important: I used the latest v6 redis (chart: release 7.6.4) as the v7 has some compatibility issues with the underlying Debian image and runs into another issue (unrelated to Gitea).

GiteaMirror commented

2025-11-02 09:11:57 -06:00

@wxiaoguang commented on GitHub (Apr 2, 2023):

You can modify your redis conn string in your Gitea config , add a "prefix" with slot, then p.c.Rename(ctx, poldsid, psid) in code won't report the CROSSSLOT error. I guess: redis://..../?...&prefix={session}:

@wxiaoguang commented on GitHub (Apr 2, 2023): You can modify your redis conn string in your Gitea config , add a "prefix" with slot, then `p.c.Rename(ctx, poldsid, psid)` in code won't report the CROSSSLOT error. I guess: `redis://..../?...&prefix={session}:`

GiteaMirror commented

2025-11-02 09:11:57 -06:00

@lunny commented on GitHub (Apr 2, 2023):

See the possible configuration connection string https://github.com/go-gitea/gitea/blob/main/modules/nosql/redis.go#L18-L22

@lunny commented on GitHub (Apr 2, 2023): See the possible configuration connection string https://github.com/go-gitea/gitea/blob/main/modules/nosql/redis.go#L18-L22

GiteaMirror commented

2025-11-02 09:11:57 -06:00

@wxiaoguang commented on GitHub (Apr 2, 2023):

The "prefix" option is not documented, but cluster needs it. Make it prefix={session}: , then all sessions keys become {session}:random-session-id, it could make the redis cluster happy.

@wxiaoguang commented on GitHub (Apr 2, 2023): The "prefix" option is not documented, but cluster needs it. Make it `prefix={session}:` , then all sessions keys become `{session}:random-session-id`, it could make the redis cluster happy.

GiteaMirror commented

2025-11-02 09:11:57 -06:00

@silverwind commented on GitHub (Apr 3, 2023):

Tangentially related: There is also redis-sentinel, a kind of "proxy" daemon that may need special support inside our code. I think it can be used in addition to redis-cluster. It is the premier HA option for redis.

@silverwind commented on GitHub (Apr 3, 2023): Tangentially related: There is also `redis-sentinel`, a kind of "proxy" daemon that may need special support inside our code. I think it can be used in addition to `redis-cluster`. It is the premier HA option for redis.

GiteaMirror commented

2025-11-02 09:11:57 -06:00

@wxiaoguang commented on GitHub (Apr 5, 2023):

To fix the problem, either:

Let users do "?prefix={session}:" trick

OR -

Do not use Rename in redis session provider , just Get and Set, then no cross-cluster problem.

@wxiaoguang commented on GitHub (Apr 5, 2023): To fix the problem, either: 1. Let users do "?prefix={session}:" trick - OR - 2. Do not use `Rename` in [redis session provider](https://github.com/go-gitea/gitea/blob/376396a08838031bb351c9ed130baf7264adb72f/modules/session/redis.go#L186) , just Get and Set, then no cross-cluster problem.

GiteaMirror commented

2025-11-02 09:11:58 -06:00

@pat-s commented on GitHub (Apr 7, 2023):

I came across this while testing redis-cluster support for the helm chart. Happy to try again once there is a built-in fix for this which does not require manual modifications. This way we can have a "clean" start when adding support for redis-cluster.

@pat-s commented on GitHub (Apr 7, 2023): I came across this while testing `redis-cluster` support for the helm chart. Happy to try again once there is a built-in fix for this which does not require manual modifications. This way we can have a "clean" start when adding support for `redis-cluster`.

GiteaMirror commented

2025-11-02 09:11:58 -06:00

@wxiaoguang commented on GitHub (Apr 7, 2023):

Can you try this PR: Use Get/Set instead of Rename when Regenerate session id #23975 ?

@wxiaoguang commented on GitHub (Apr 7, 2023): Can you try this PR: Use Get/Set instead of Rename when Regenerate session id #23975 ?

GiteaMirror commented

2025-11-02 09:11:58 -06:00

@pat-s commented on GitHub (Apr 7, 2023):

Not as a PR as I would need an image to test it with the Helm chart.

@pat-s commented on GitHub (Apr 7, 2023): Not as a PR as I would need an image to test it with the Helm chart.

GiteaMirror commented

2025-11-02 09:11:58 -06:00

@wxiaoguang commented on GitHub (Apr 7, 2023):

Approve and merge, then get a dev image?

@wxiaoguang commented on GitHub (Apr 7, 2023): Approve and merge, then get a dev image?

GiteaMirror commented

2025-11-02 09:11:58 -06:00

@lunny commented on GitHub (Apr 7, 2023):

It should be a bug rather than a feature because we have already supports redis cluster.

@lunny commented on GitHub (Apr 7, 2023): It should be a bug rather than a feature because we have already supports redis cluster.

GiteaMirror commented

2025-11-02 09:11:59 -06:00

@pat-s commented on GitHub (Apr 13, 2023):

This PR together with #24114 resulted in a successful connection to a redis-cluster instance 🎉

@pat-s commented on GitHub (Apr 13, 2023): This PR together with #24114 resulted in a successful connection to a `redis-cluster` instance 🎉

Sign in to join this conversation.

Branches Tags

1 Participants

Notifications

Due Date

No due date set.

Dependencies

No dependencies set.

Reference: github-starred/gitea#10583

Support for redis-cluster #10583

Feature Description

Screenshots

Support for `redis-cluster` #10583