kayac / gunfish Goto Github PK

It is too short period to retry, so should expands that period.

FCM recommends migrate old HTTP API to HTTP v1 API.
https://firebase.google.com/docs/cloud-messaging/migrate-v1

Values for the APNs JSON reason key have changed and thus we should update ErrorMessage const in github.com/kayac/Gunfish/apns/error.go.

Previous documentation(web.archive.org because that page cannot be found.):
https://web.archive.org/web/20160902182330/https://developer.apple.com/library/ios/documentation/NetworkingInternet/Conceptual/RemoteNotificationsPG/Chapters/APNsProviderAPI.html

Current documentation:
https://developer.apple.com/library/content/documentation/NetworkingInternet/Conceptual/RemoteNotificationsPG/CommunicatingwithAPNs.html

When many error occur, It is often high cpu usage.
So, some command workers spawn in advance and continue to use own stdin until died.

Sending a request like the one below fails to set the Header parameters sent to the APNS server

[{
  "header": {
    "apns-topic": "com.kayac.mypackage"
  },
  "token": "XXXXXXX",
  "payload": {
    "aps": {
      "alert": {
        "body": "test"
      }
    }
  }
}]

The response in the error hook clearly shows that the header has not been set:

{"response":{"apns-id":"","status":400},"response_time":0.371444302,"request":{"header":{},"token":"XXXXXXX","payload":{"aps":{"alert":{"body":"test"}}},"tries":0},"error_msg":{"reason":"MissingTopic","timestamp":0},"resp_uid":"aaaaaaaaa"}

Gunfish doesn't retry when some FCM v1 errors occur.

When FCM returns INTERNAL or UNAVAILABLE error, it should be retried.
https://firebase.google.com/docs/cloud-messaging/http-server-ref?hl=ja#error-codes

{
    "file": "/home/runner/work/Gunfish/Gunfish/supervisor.go",
    "level": "error",
    "line": "432",
    "msg": "Unknown error message: INTERNAL",
    "res_queue_size": 0,
    "resend_cnt": 0,
    "resp_uid": "6d9db339-bcdf-4db5-9b64-14b671e978ec",
    "response_time": 5.047958725,
    "time": "2024-03-26T13:07:19Z",
    "token": "xxx",
    "type": "worker",
    "worker_id": 61
}

https://firebase.google.com/docs/cloud-messaging/migrate-v1

Apps using the deprecated FCM legacy APIs for HTTP and XMPP should migrate to the HTTP v1 API at the earliest opportunity. Sending messages (including upstream messages) with those APIs was deprecated on June 20, 2023, and will be removed in June 2024.

FCM Legacy API will be removed soon. Gunfish will drop supports the legacy API (/push/fcm).

FCM returns NotRegistered, but Gunfish can't handle the errors.

Google FCMv1 endpoint returns 'NOT_FOUND' error message if token is replaced. So the error hook should be invoked to remove the old token.

Steps to reproduce:

1. Registered a device (and successfully sent a message via FCMv1 gateway)
2. Deleted app and reregistered device (results in new token, with the first beeing still stored)
3. Send message to the old (and now orphan) token

Result:

time="2020-01-23T09:25:02Z" level=error msg="Unknown error message: NOT_FOUND" file=/home/circleci/project/supervisor.go line=432 res_queue_size=0 resend_cnt=0 resp_uid=07cd1248-7ee4-4d9c-9169-894cce461274 response_time=0.3715487 token="[redacted]" type=worker worker_id=0

(Tested with v0.4.1-rc2 Docker container: https://hub.docker.com/r/kayac/gunfish/tags?page=1&ordering=last_updated)

To me it looks like FCM is responding something like this, where the status is NOT_FOUND but the details errorCode is UNREGISTERED:

{
  "error": {
    "code": 404,
    "message": "Requested entity was not found.",
    "status": "NOT_FOUND",
    "details": [
      {
        "@type": "type.googleapis.com/google.firebase.fcm.v1.FcmError",
        "errorCode": "UNREGISTERED"
      }
    ]
  }
}

This is also discussed in similar projects facing the same issue:

• https://stackoverflow.com/questions/55689804/firebase-cloud-messaging-tokens-stop-working-404-not-found
• https://stackoverflow.com/questions/48928972/fcm-http-v1-api-returns-404-for-unregistered-token

I don't want to invoke a command by error_hook, because my APNS tokens list includes too many invalid tokens.
I hope to run Gunfish with no error_hook, and I will scan the Gunfish logs and find invalid tokens from that logs at first.