native service delete errors for old allocs after client restart

[mr-karan]

Nomad version

1.7.7

Operating system and Environment details

• Running in an AWS environment
• Ubuntu 24.04

Issue

Service registration errors and task failures occurring during node registration.

Reproduction steps

1. Node starts registration process
2. Multiple service registration deletion attempts fail
3. Template rendering issues occur for HAProxy peer service
4. Sibling task failures cascade to other services

Expected Result

• Clean node registration
• Successful service registration management
• Proper template rendering for HAProxy peer service
• Successful task execution without cascading failures

Actual Result

Multiple cascading failures observed:

1. Service registration errors:

[ERROR] client.rpc: error performing RPC to server: error="rpc error: rpc error: service registration not found"

2. Template failures:

Missing: nomad.service(haproxy-peer)

3. Task failures:

Setup Failure: failed to setup alloc: pre-run hook "group_services" failed: no servers

4. Forced termination:

Exit Code: 0, Exit Message: "executor: error waiting on process: rpc error: code = Canceled desc = grpc: the client connection is closing"

Nomad Client logs

Nov 14 08:11:51 [INFO]  agent: (runner) starting
Nov 14 08:11:51 [ERROR] client.rpc: error performing RPC to server: error="rpc error: rpc error: service registration not found" rpc=ServiceRegistration.DeleteByID server=172.31.2.217:4647
Nov 14 08:11:51 [INFO]  client.service_registration.nomad: attempted to delete non-existent service registration: service_id=_nomad-task-d90c47ce-f4be-0fa3-e019-5d1b522e64a1-group-haproxy-default-haproxy-peer-haproxy-peer-net namespace=kite
Nov 14 08:12:01 [INFO]  client: node registration complete

Nomad Alloc Events Timeline

Nov 14, '24 08:10:36 - Terminated (Exit Code: 0)
Nov 14, '24 08:10:35 - Killing (Sent interrupt, 5s grace period)
Nov 14, '24 08:10:33 - Template Missing: nomad.service(haproxy-peer)
Nov 14, '24 08:10:30 - Sibling Task Failed (prepare-logging-setup)
Nov 14, '24 08:10:30 - Setup Failure (group_services hook failed)
Nov 14, '24 07:31:19 - Started

The primary issue appears to be related to service registration and template rendering failures, particularly affecting HAProxy peer services. This is causing cascading failures across dependent services and tasks.

[tgross]

@mr-karan if the node hasn't registered yet, how is it running services? Is this a node that was running services and then restarted?

[tgross]

@mr-karan we haven't heard back on this one in a while. It looks like you've got some running jobs and then you're rebooting the client agent, and then the allocations fail to restore? Are you rebooting the node? I tried reproducing but there really isn't enough to go on. I'm going to close this out for now as unreproducible, but if you have more info I'd be happy to reopen.

[tgross]

Reproduced!

jobspec

job "httpd" {

  group "web" {

    network {
      mode = "bridge"
      port "www" {
        to = 8001
      }
    }

    service {
      name     = "httpd-web"
      provider = "nomad"
      port     = "www"
    }

    task "http" {

      driver = "docker"

      config {
        image   = "busybox:1"
        command = "httpd"
        args    = ["-vv", "-f", "-p", "8001", "-h", "/local"]
        ports   = ["www"]
      }

      identity {
        env  = true
        file = true
      }

      resources {
        cpu    = 100
        memory = 100
      }

    }
  }
}

Running on a cluster with a single client and single server, I was running that job for a while and updating it frequently debugging other work. Then I see the following logs on the client after running systemd restart nomad:

2024-12-12T16:38:44.311-0500 [ERROR] client.rpc: error performing RPC to server: error="rpc error: service registration not found" rpc=ServiceRegistration.DeleteByID server=10.37.105.3:4647
2024-12-12T16:38:44.311-0500 [ERROR] client.rpc: error performing RPC to server which is not safe to automatically retry: error="rpc error: service registration not found" rpc=ServiceRegistration.DeleteByID server=10.37.105.3:4647
2024-12-12T16:38:44.311-0500 [INFO] client.service_registration.nomad: attempted to delete non-existent service registration: service_id=_nomad-task-c1c2f114-5f48-bc51-3575-95764f088e89-group-web-httpd-web-www namespace=default
2024-12-12T16:38:44.339-0500 [ERROR] client.rpc: error performing RPC to server: error="rpc error: service registration not found" rpc=ServiceRegistration.DeleteByID server=10.37.105.3:4647
2024-12-12T16:38:44.340-0500 [ERROR] client.rpc: error performing RPC to server which is not safe to automatically retry: error="rpc error: service registration not found" rpc=ServiceRegistration.DeleteByID server=10.37.105.3:4647
2024-12-12T16:38:44.340-0500 [INFO] client.service_registration.nomad: attempted to delete non-existent service registration: service_id=_nomad-task-1bec7c14-86a2-70e6-c018-654a05d02cc4-group-web-example-web-www namespace=default
2024-12-12T16:38:44.423-0500 [ERROR] client.rpc: error performing RPC to server: error="rpc error: service registration not found" rpc=ServiceRegistration.DeleteByID server=10.37.105.3:4647
2024-12-12T16:38:44.423-0500 [ERROR] client.rpc: error performing RPC to server which is not safe to automatically retry: error="rpc error: service registration not found" rpc=ServiceRegistration.DeleteByID server=10.37.105.3:4647
2024-12-12T16:38:44.424-0500 [INFO] client.service_registration.nomad: attempted to delete non-existent service registration: service_id=_nomad-task-1e098779-8d6e-8840-372f-178d44ca90c6-group-web-httpd-web-www namespace=default

The allocation IDs here are all for the service registration of the old allocations, not the ones that exist currently. So the errors we get from the server make sense -- these should all be gone already. But why the client still thinks it has to delete them I don't know yet. Seems like a chunk of data is getting left behind in the client state store.

Reopening and marking for roadmapping.