Stacked on #14580. Review and merge that first; this PR then retargets to main.

What does this PR do?

createPush({ users: [...] }) reaches a user who has no push target through Appwrite push on users/<userId>, so an app can receive direct messages without registering a target. Until now that no-target fallback in streamRecipients ran only when $default was Appwrite. $default is findOne(providers, enabled, type = push), in practice the first push provider created. A project that set up FCM or APNs before Appwrite push got No valid recipients found. for those users, even with an Appwrite provider enabled.

The default provider doesn't send by itself. A user with no targets has no FCM or APNs token, so in that case nothing reached them at all.

Fix

• checkAppwritePushEnabled() becomes getAppwritePushProvider(). It returns the enabled Appwrite push provider (the default when that's Appwrite), resolved once per send and seeded into the provider cache.
• The targetless fallback yields that provider's page, whatever the default is.
• The wake-up flag uses the same lookup, so its behaviour is unchanged.

The project's default provider is left alone on purpose. Targets saved without a providerId follow the default, and making Appwrite the default would reroute their FCM and APNs tokens to MQTT.

message.providers only lists providers the send used

The worker built the message.providers span, and its own appended search text, from $providers, the resolved-provider cache. That cache is seeded with the default provider, and now with the enabled Appwrite provider. So a send delivered entirely through another provider still reported providers it never touched.

sendBatch() now returns the provider it ran on. The worker collects those providers from page batches and wake-up batches, and builds the span from that set. The cache keeps its seeding, so there are no extra queries.

The message search field itself was never affected. The messageSearch attribute filter (app/init/database/filters.php) recomputes it on every write from the ID, description, status, title and type, so the worker's appended provider names are overwritten on save. That append loop is effectively dead code and is left as is here.

Matrix: user sends with Appwrite + FCM enabled

Users holding an FCM or APNs target but no Appwrite target are deliberately left alone. They already get the full notification, and adding MQTT would deliver it twice.

Trade-off

For each targetless user, a send:

• provisions their users/<userId> topic row on first use;
• writes a pushLedger row (kept for 7 days for replay);
• counts toward deliveredTotal and messages.sent, whether or not a device is connected.

This already happened when Appwrite was the default. It now applies whenever an Appwrite provider is enabled. We keep it on purpose: receiving on users/<userId> with no target setup is a deliberate convenience.

Test Plan

E2E in tests/e2e/Services/Mqtt/MqttServerTest.php, each in a fresh project so FCM really is the first enabled push provider:

• testUserTopicReceivesPushWhenAnotherProviderIsDefault: FCM is created first, then Appwrite. A targetless user receives the message on users/<userId> and the message ends sent.
• testUserPushWithoutEnabledAppwriteProviderFails: FCM is enabled and Appwrite disabled. A targetless user's message ends failed.

The FCM provider uses placeholder credentials. The user holds no FCM token, so Google is never called.

The Mqtt CI lane mapping and the rust-toolchain repin come from #14580.

Locally I ran php -l, Pint and PHPStan on the changed files. PHPStan reports no new errors; the only ones are the Utopia\Database\Id not-found errors that main also shows because the local vendor is stale. E2E runs in CI.

Related PRs and Issues

• Stacked on fix(messaging): deliver Appwrite push topic campaigns on the topic id #14580, which fixes topic sends on the topic ID.

🤖 Generated with Claude Code