will
bcf8df5579
Read knadh/listmonk's actual Go source (cmd/campaigns.go, cmd/subscribers.go) instead of relying on incomplete public docs. Two corrections: the media attach field is "media" (plain IDs), not "media_ids" as originally guessed; and there's a single query-based bulk list action (PUT /api/subscribers/query/lists) that applies a segment_query server-side, so resolveSegment no longer fetches subscriber IDs client-side before bulk-adding them. Create-status-defaults-to-draft and the test-send "subscribers" field are now confirmed rather than flagged TODO(verify).
107 lines
8.8 KiB
Markdown
107 lines
8.8 KiB
Markdown
# eec-campaigns
|
|
|
|
Broadcast/segment email, authored as Markdown+YAML in git, driving [listmonk](https://listmonk.app)'s real Campaign API — not [`eec`](https://gitea.reground.org/will/eec)'s transactional `/api/tx` path, which has no unsubscribe or bulk-send machinery. Push to `master` and CI syncs `campaigns/` straight to listmonk as **draft** campaigns; a separate, deliberate action (a pushed tag or a manual workflow run) is what actually sends one. A push here can never blast your list by itself.
|
|
|
|
## Layout
|
|
|
|
```
|
|
campaigns/
|
|
<slug>/
|
|
campaign.md
|
|
assets/
|
|
one-pager.pdf
|
|
```
|
|
|
|
Each campaign is a directory under `campaigns/`, named for its slug. **The directory name is the campaign's identity** — it's sent to listmonk as the campaign's `name` and is what `sync`/`send`/`test` all look it up by. There's no separate `name:`/`slug:` field in frontmatter to keep in sync with the directory; renaming the directory creates a new campaign in listmonk rather than renaming the existing one.
|
|
|
|
## `campaign.md`
|
|
|
|
```yaml
|
|
---
|
|
subject: "Big Announcement"
|
|
lists: ["Newsletter"]
|
|
from_email: hello@reground.org
|
|
tags: ["announcement"]
|
|
template_id: 4
|
|
segment_query: >-
|
|
subscribers.attribs->>'source' = 'workshop-signup'
|
|
AND subscribers.attribs->>'eec_enroll' IS NULL
|
|
preview_emails:
|
|
- cofounder@reground.org
|
|
attachments:
|
|
- assets/one-pager.pdf
|
|
---
|
|
|
|
Campaign body goes here, in Markdown. It's sent to listmonk with
|
|
content_type "markdown" — listmonk renders the HTML *and* derives the
|
|
plaintext alternative itself, so there's no separate plain-text copy to
|
|
maintain.
|
|
```
|
|
|
|
- **`subject`** — required, the email subject line.
|
|
- **`lists`** — listmonk list **names** (not numeric IDs, unlike `eec`'s `course_steps.list_id`), resolved to IDs at sync time. At least one of `lists`/`segment_query` is required — a campaign needs a target audience.
|
|
- **`from_email`** — required.
|
|
- **`tags`**, **`template_id`** — optional; `template_id` overrides listmonk's default template (see [Before your first real send](#before-your-first-real-send)).
|
|
- **`segment_query`** — optional, see [Segmentation](#segmentation).
|
|
- **`preview_emails`** — optional, overrides the default preview address(es) for this campaign's automatic post-sync preview (see [Automatic preview](#automatic-preview)).
|
|
- **`attachments`** — optional, paths relative to the campaign's own directory, uploaded to listmonk's media library and attached to the campaign.
|
|
- **`type`** — defaults to `regular`. `optin` campaigns aren't supported (no confirmation workflow here) and are rejected at sync time.
|
|
- **`scheduled_at` doesn't exist** — deliberately. listmonk auto-sends a `scheduled` campaign the moment it fires, which would let a plain `git push` cause a real send. Scheduled sends need their own explicit, still manually-triggered flow if ever wanted.
|
|
|
|
## Segmentation
|
|
|
|
`segment_query` is a raw Postgres-style SQL boolean expression — the same segmentation mechanism listmonk's own admin UI search box already accepts (`AND`/`OR`/`NOT`, `IN`, `LIKE`, `EXISTS` subqueries against e.g. `campaign_views` for engagement-based segments). No custom query language here; the string is passed straight through to listmonk's query-based bulk list action.
|
|
|
|
Since a listmonk campaign can only target whole list(s), not an arbitrary query directly, `sync` materializes the query into a managed list on your behalf:
|
|
|
|
1. Find or create a list named `segment:<slug>`.
|
|
2. Ask listmonk to add every subscriber matching the expression to that list, in one call — no subscriber IDs are ever fetched client-side, listmonk applies the query and the list membership change server-side.
|
|
3. Target the campaign at that list, alongside any named `lists:` also given.
|
|
|
|
That list is a point-in-time snapshot taken at sync time, not a live dynamic segment — push again before sending to pick up newly-matching subscribers. An invalid SQL fragment fails that campaign's sync with listmonk's own error message; nothing is created or sent.
|
|
|
|
## Attachments
|
|
|
|
Each file in `attachments:` is uploaded to listmonk's media library under a synthesized name (`<slug>-<content-hash>-<original filename>`), so editing the file produces a fresh upload while an unchanged file is recognized and reused — there's no local manifest of what's already been uploaded, listmonk's own media library is the only state this tool depends on.
|
|
|
|
## Automatic preview
|
|
|
|
Every `sync` that actually creates or changes a campaign (re-syncing identical content is a no-op — no API write, no preview) automatically sends a preview via listmonk's test-send endpoint, to `preview_emails` if the campaign sets it, otherwise to the repo-wide default (`CAMPAIGNS_PREVIEW_EMAIL`, a CI secret). Check your inbox after pushing — there's no separate manual step for the common "does this look right" check. `campaigns test <slug> <email>` (or the `Test-send campaign` workflow) is there for an on-demand re-preview without touching content.
|
|
|
|
## Sending
|
|
|
|
Sync only ever produces a **draft**. Two ways to actually send one:
|
|
|
|
- **Push a tag** — `git tag send/<slug> && git push origin send/<slug>`. This is the primary, git-native trigger: a durable, auditable record (`git log --tags` shows every real send that ever happened). These are "command tags," not release tags — delete and re-push if a send needs retrying; the real safety net is listmonk's own status check (`send` only works from `draft`/`paused`), not tag uniqueness.
|
|
- **Manually run the `Send campaign` workflow** in Gitea Actions, entering the slug — for when a button is handier than a local tag push.
|
|
|
|
Both run the exact same `campaigns send <slug>` command.
|
|
|
|
## Sync guardrails
|
|
|
|
A rejected campaign (check the CI run for `rejected:`) is usually one of these, all deliberate:
|
|
|
|
- **The campaign is anything other than `draft` in listmonk** (`scheduled`/`running`/`paused`/`cancelled`/`finished`) — sync refuses to touch it. Once something's live or sent, git no longer has write access to it.
|
|
- **A list name in `lists:` matches zero or more than one listmonk list** — sync never guesses which one you meant.
|
|
- **An invalid `segment_query`** — listmonk's own query error is surfaced verbatim.
|
|
- **Missing `subject`/`from_email`/a target audience**, or `type: optin`.
|
|
|
|
One bad campaign in the push doesn't block the others — check `rejected:` in the sync job's log for which ones and why.
|
|
|
|
## Before your first real send
|
|
|
|
listmonk's only currently-provisioned email template (`eec-passthrough`) is a bare passthrough used for `eec`'s transactional course emails — no unsubscribe footer, no branding. Create (or confirm) a real campaign template with a proper unsubscribe link in the listmonk admin UI before running your first `send`; drafts and previews render fine without one, but a real send without it isn't compliant.
|
|
|
|
## Open items
|
|
|
|
Every request shape in `internal/listmonk` is now confirmed against `knadh/listmonk`'s actual Go source (not just its docs, which are incomplete on a few of these): the campaign create/update payload, its `media` field for attachments (listmonk's request/response asymmetry — requests send plain IDs under `media`, responses echo full objects back under the same key), create always defaulting to `draft` regardless of any caller-supplied status, the test-send endpoint's `subscribers` field, and the query-based bulk list action (`PUT /api/subscribers/query/lists`) segmentation uses. What's left is genuinely operational, not code:
|
|
|
|
- **A real, unsubscribe-capable campaign template** needs to exist in the live listmonk instance before a real send (see above) — a content/admin-UI task, not something this tool verifies for you.
|
|
- **Provisioning the dedicated listmonk API user and pasting its token into this repo's Gitea secrets** (see Deploying below) hasn't happened yet.
|
|
- **Gitea Actions' tag-push and `workflow_dispatch` triggers** are standard, long-supported Actions syntax and the runner already successfully uses `uses: actions/checkout@v4` elsewhere in this org, so this should work as written — but it's still worth confirming the first time `send.yml` actually fires.
|
|
- The source was read against `knadh/listmonk`'s `master` branch; if the live instance runs a substantially older or newer version, a quick diff against its own `cmd/campaigns.go`/`cmd/subscribers.go` is cheap insurance before the first real send.
|
|
|
|
## Deploying
|
|
|
|
Push to `master` → `.gitea/workflows/sync.yml` runs `campaigns sync .` directly against listmonk (`CAMPAIGNS_LISTMONK_BASE_URL`/`_API_USER`/`_API_TOKEN`, Gitea Actions secrets). No separate deploy step or intermediate service — this tool has no server or database of its own, listmonk is the only state store. Provisioning the dedicated listmonk API user follows `eec`'s existing Ansible role pattern; pasting the resulting token into this repo's Gitea secrets is a manual one-time step (nothing automates writing Gitea repo secrets today).
|