Second of two changes for #186; closes it. Each kind of object the console serializes on now has a family, and every advisory key is built one way: hashtextextended('<family>:<id>', 0), with a UUID written in canonical form. db.Lock and db.LockShared (internal/db/locks.go) build the key and take the lock on a *sql.Tx only, and each family's statement begins with a name line, so pg_stat_activity and the test traces show which family a backend waits on. Families, in the order a transaction takes them: materialization, domains-registry, fedwiki-workspace, lifecycle, stripe-subscription, stripe-invoice, stripe-product-sync, desired-state. The lifecycle and desired-state keys are unchanged byte for byte; the other six change. Migration 00040 replaces core.assert_rendezvous to look for the new materialization key (owner, grants, SECURITY DEFINER and search_path kept; Down restores migration 18's body), and both seeds take the new key right after BEGIN. docs/database-locks.md lists the families in a table and writes one order every path keeps: the materialization lock first, then the families in table order, then rows. The project lint gains advisory-lock-site (an advisory lock function named in a Go string or in SQL or shell code outside the helpers, the test kit's probe, the migrations that name the rendezvous and the rule's own tests; a seed taking any other key, or the rendezvous anywhere but first after BEGIN), pool-lock-site (a core.resource_pools row lock outside LockResourcePool and LockGrantPool, or a call or method value of those outside entitlements/tx.go) and lock-families (the doc's table against the code). The walk covers cmd/, scripts/, internal/, test/ and the root's Go files, and skips testdata and directories whose name begins with a dot, as the go tool does. Tests take a family's lock through the same helpers and probe one with dbtest.TryLock. Recording a lifecycle request comes in two forms: RecordLifecycleRequest opens its own transaction on a database or a pinned connection, and RecordLifecycleRequestInTx runs in the caller's *sql.Tx, so a connection is never used outside a transaction (which released the instance lock at once) and does not compile in the second. The same holds for RecordFirstLifecycleRequest. docs/production-deployment.md now states the upgrade order for every upgrade: stop every process of the running release, run migrate up if migrations run separately, then start the new release; two releases running at once can build different keys for one lock and share none. docs/go-conventions.md, docs/testing.md and docs/building-an-integration.md follow the code. The lint command's help text names the files the lock rules read. 202 golden lines in 109 files change their statement text. Deploy: stop every process of the previous release first. Rollback: stop this release; as the migration user, in one transaction, run 00040's Down section and DELETE FROM goose_db_version_core WHERE version_id = 40; start the previous release (migrate down rolls back the last integration stream first). A deployment using console_reader re-runs scripts/db-reader-role.sql. Break list: 35 records, 35 CAUGHT. Named without a record: the *sql.Tx parameters of the helpers and the -InTx forms (a break that passes a connection does not compile, and the build is the check), and the comments. Full suite on reset databases: 5727 passed, 16 skipped, none failed.
Temporal Workflows
This directory contains the Temporal workflow definitions, activity implementations, and worker configuration for the member-console application.
Structure: Domain-Based
We use a Domain-Based (or Feature-Based) structure. Instead of grouping all workflows in one folder and all activities in another, we group them by business domain.
internal/workflows/
├── client.go # Temporal client factory
├── worker.go # Central worker setup & registration
├── queues/
│ └── queues.go # Task Queue constants
├── common/ # Shared utilities & generic activities
│ └── options.go # Standard ActivityOptions (retries, timeouts)
├── example/ # "Example" Domain
│ ├── workflow.go # Workflow definition
│ └── activities.go # Domain-specific activities
├── fedwiki/ # "FedWiki" Domain - Site provisioning
│ ├── README.md # Domain documentation
│ ├── FARMMANAGER.md # FarmManager API documentation
│ ├── workflow.go # Create/Delete site workflows
│ ├── activities.go # Site management activities
│ └── farmmanager_client.go # HTTP client for FarmManager API
└── onboarding/ # (Future) "Onboarding" Domain
├── workflow.go
└── activities.go
Adding a New Domain
- Create a directory:
internal/workflows/<domain_name>/ - Define the Workflow: Create
workflow.goin that directory. - Define Activities: Create
activities.goin that directory.- Define a struct (e.g.,
Activities) to hold dependencies (DB, Logger). - Implement a
NewActivitiesconstructor.
- Define a struct (e.g.,
- Register: Go to
internal/workflows/worker.goand add your new domain to theNewWorkerfunction:
// internal/workflows/worker.go
// ... inside NewWorker ...
w.RegisterWorkflow(yourdomain.Workflow)
w.RegisterActivity(yourdomain.NewActivities(cfg.DB, cfg.Logger))
Usage
Starting the Worker
The worker is embedded in the main application but can be extracted later.
import "git.coopcloud.tech/wiki-cafe/member-console/internal/workflows"
// ... in your main setup ...
workerCfg := workflows.DefaultWorkerConfig(database, logger)
worker, err := workflows.NewWorker(temporalClient, workerCfg)
if err := worker.Start(); err != nil {
// handle error
}
defer worker.Stop()
Starting a Workflow
import (
"git.coopcloud.tech/wiki-cafe/member-console/internal/workflows"
"git.coopcloud.tech/wiki-cafe/member-console/internal/workflows/queues"
"git.coopcloud.tech/wiki-cafe/member-console/internal/workflows/example"
)
// ... inside an HTTP handler ...
workflowOpts := client.StartWorkflowOptions{
ID: "example-" + uuid.New().String(),
TaskQueue: queues.Main,
}
we, err := temporalClient.ExecuteWorkflow(
ctx,
workflowOpts,
example.Workflow, // Pass the function reference
example.Input{Message: "hello"},
)