Trigger

./infrastructure/session.sh check exits 1 at the end of a session, or before the start of one.

Impact

The session layer bills until it is destroyed (ADR 0009). A cluster that remains also keeps vCPU quota in use, and the regional quota is 4. The next cluster cannot be created while it remains (docs/architecture.md, risk table).

First checks

1. ./infrastructure/session.sh check. It lists every resource group with its resource counts, then prints one line per category, none or REMAINS. Exit 1 means something remains. Exit 0 means nothing remains. Exit 2 means an error or wrong use, for example the Azure CLI is not signed in. The output is masked and shows names and counts, not IDs.
2. Read the == result lines. check reports:
   • the session resource group pp-dev-session-rg or the node resource group pp-dev-aks-nodes-rg,
   • a cluster, virtual machine or scale set,
   • a public IP address, a load balancer or a managed disk,
   • a resource group that is not pp-shared-state-rg, pp-dev-persistent-rg, NetworkWatcherRG or MA_pp-dev-amw_swedencentral_managed (made by Azure for the Monitor workspace).
3. ./infrastructure/session.sh down. It shows the destroy plan and destroys nothing. It shows what Terraform would still remove.

Likely causes

• down --yes was not run, or it stopped with an error before it finished.
• The destroy failed part way. Terraform keeps what remains in the state.
• The destroy stopped on a public IP that the load balancer still holds (PublicIPAddressCannotBeDeleted). Azure refuses to delete the IP while the cluster is being deleted. down --yes now repeats once after a wait. If both attempts fail, run it again.
• A state lock error (“Error acquiring the state lock”) stops the command. The cause is not confirmed (infrastructure/README.md).
• A resource exists in Azure that is not in the Terraform state. The repository has no procedure for this case. It is not known what would create one.

Other causes are not known from the repository.

Fix

1. ./infrastructure/session.sh down. Read the plan. It shows what will be destroyed.
2. ./infrastructure/session.sh down --yes. It destroys the session root. down needs no authorized range and no admin group value.
3. If it stops with a state lock error, run the command again. Whether that clears the error is not known. If it stops on a public IP (PublicIPAddressCannotBeDeleted), run it again. A second run cleared this error in Session 1.
4. ./infrastructure/session.sh check. Repeat until it exits 0.
5. If something remains that Terraform does not know, the repository gives no way to remove it. Look at the names in the check output, decide by hand, and run check again. Do not remove pp-shared-state-rg, pp-dev-persistent-rg, NetworkWatcherRG or MA_pp-dev-amw_swedencentral_managed: check does not count them, and the first two hold state and persistent resources.

Verify

1. ./infrastructure/session.sh check. It exits 0, and the result lines show none. The note about pp-dev-persistent-rg is expected.

Rollback

A destroy cannot be undone. The session layer is created again with ./infrastructure/session.sh up and up --apply (ADR 0009).

Source

infrastructure/README.md (what check lists), infrastructure/session.sh, ADR 0009, Session 1 and Session 2 teardown steps.

Status

not exercised