@devries-cameron20 / t3code Mirror · read-only
Fork 0

The best way to control Claude Code, Codex and more from any computer or phone

Mirrored from https://github.com/Camo67/t3code.git

Synced Nov 21, 07:15 UTC

main
md 67 lines 3.3 KB
Rendered Raw
1 # Nightly batch on a Cloud Run Job
2
3 A Cloud Run Job that rolls the last 24 hours of orders in BigQuery up into a JSON summary in Cloud Storage. Cloud Scheduler runs it every night, and a public admin route runs it on demand.
4
5 - [`Summarize`](./src/Summarize.ts): a `GCP.Run.Job` with an Effect-native `run` entry. It queries the `orders` table grouped by region and writes `daily/<YYYY-MM-DD>.json` with per-region and overall order counts and revenue. Re-running on the same day overwrites that day's object.
6 - [`Nightly`](./src/Nightly.ts): a `GCP.CloudScheduler.Job` that `POST`s to the Cloud Run Admin API's `jobs:run` endpoint for `Summarize` at 02:00 UTC. It authenticates with an OAuth token for the job's own runtime service account, which a `GCP.IAM.Member` grants `roles/run.invoker` on the job.
7 - [`Admin`](./src/Admin.ts): a public `GCP.Function` (Cloud Run service). `POST /run` starts an execution and returns its operation name.
8 - [`Orders`, `Reports`](./src/resources.ts): the BigQuery table (in the `Warehouse` dataset) and the bucket the summaries land in.
9
10 A summary looks like this:
11
12 ```json
13 {
14 "from": "2026-09-26T02:00:04.120Z",
15 "to": "2026-09-27T02:00:04.120Z",
16 "execution": "summarize-abcde",
17 "orders": 5,
18 "revenueCents": 5999,
19 "regions": [
20 { "region": "east", "orders": 3, "revenueCents": 3999 },
21 { "region": "west", "orders": 2, "revenueCents": 2000 }
22 ]
23 }
24 ```
25
26 ## Bindings and IAM
27
28 | Binding / resource | Grants |
29 | --- | --- |
30 | `GCP.BigQuery.ReadTable` + `ReadTableHttp` (on `Summarize`) | the job's runtime account `roles/bigquery.dataViewer` on the table and `roles/bigquery.jobUser` on the project, because query jobs can only be granted at project level |
31 | `GCP.Storage.WriteBucket` + `WriteBucketHttp` (on `Summarize`) | the job's runtime account `roles/storage.objectUser` on the bucket |
32 | `GCP.Run.RunJob` + `RunJobHttp` (on `Admin`) | the service's runtime account `roles/run.jobsExecutorWithOverrides` on the job |
33 | `GCP.IAM.Member("NightlyRunner")` | the job's runtime account `roles/run.invoker` on the job, which includes `run.jobs.run`. Cloud Scheduler mints the OAuth token for that account. |
34
35 Cloud Scheduler uses an OAuth token, not OIDC, because the target is a `*.googleapis.com` API rather than a Cloud Run URL. `Admin` sets `invokerIamDisabled: true` so the example can be driven with `curl`; in production, drop it and call `/run` with an identity token.
36
37 ## Deploy
38
39 From the repository root:
40
41 ```sh
42 pnpm install
43 cd examples/gcp-scheduled-job
44 pnpm deploy --profile <name>
45 ```
46
47 This needs a local Docker daemon to build the images. Then start a run with the printed `url`:
48
49 ```sh
50 curl -X POST "$URL/run"
51 ```
52
53 A Cloud Run Job execution can take a couple of minutes to start. To run the schedule itself, use `gcloud scheduler jobs run <schedulerJobName> --location <region>`.
54
55 ## Live test
56
57 ```sh
58 ALCHEMY_PROFILE=<name> bun test
59 ```
60
61 The test deploys the stack and seeds five orders with BigQuery `insertAll`. It calls `POST /run`, polls the bucket until a summary with the seeded regions appears, and checks the counts and totals. It then forces the Cloud Scheduler job with `cloudscheduler.runProjectsLocationsJobs` and waits for a new execution of the job. Finally it destroys the stack and verifies the job, the scheduler job, and the bucket are gone.
62
63 ## Destroy
64
65 ```sh
66 pnpm destroy --profile <name>
67 ```