Jobs
Este conteúdo não está disponível em sua língua ainda.
Every job type the server runs and every job it starts on a schedule: what the job does, what starts it, how it retries, and what running it
again by hand does. Use this page to read the types on the admin jobs page, in my job list, and in the my.job.* metrics. How the queue runs, and the commands and pages that manage it, are in
The job queue and failed (dead) jobs.
The server has 110 job types. 46 periodic jobs start some of them on a schedule; Periodic jobs lists those.
A worker can run only some of these types: Splitting workers by job type.
Reading the tables
Section titled “Reading the tables”Started by says what puts the job in the queue. A schedule is listed under Periodic jobs.
On failure says what the job does when it cannot finish:
| Value | Meaning |
|---|---|
| Default | Up to 10 runs. The wait starts at 30 seconds and doubles after each failed run, up to 24 hours, with up to 20% taken off at random. After the last run the job becomes a dead job. |
| Transactional mail | Up to 6 runs over about 15 minutes, since a late link is of little use. |
| The default retries for a temporary failure. A permanent rejection, or a message that is not valid, makes it a dead job at once. |
For every type that has a payload, a payload that cannot be read makes the job dead at once: running it again fails the same way.
Run again says what my job retry, or Retry on the jobs page, does with a dead job of the type. All job handlers are written to be repeated, since a job can run more than once anyway, but what repeating means varies:
| Value | Meaning |
|---|---|
| Safe | Work that is already done is not done again. |
| Next run | Safe, and rarely needed: the next scheduled run does the same work. Delete the dead job once you have fixed the cause. |
| Sends again | Safe for this server, but a message that already went out can go out again. A mail server or push service offers no way to send a message only once, so a run that stops after the server took the message and before the job finished sends it a second time. |
A job type can also be registered or do its work only with some settings; Needs in the tables and Jobs that depend on settings say which.
Federation
Section titled “Federation”| Job type | What it does | Started by | On failure | Run again |
|---|---|---|---|---|
federation.inbox |
Processes one activity received from another server. Activities of one actor run in order. | The inbox accepting an activity | Default. An activity that can never be processed is dead at once. | Safe. What it already recorded stays as it is, and an answer this server sends, such as the Reject of a refused Follow, goes out once. An Accept or Reject that does not name a known follow applies to the follow between the two accounts now. After 7 days the activity’s document is gone and the job is dead at once. |
federation.fanout |
Works out the inboxes of one outgoing activity, one batch at a time, and queues a federation.deliver for each new delivery. Activities whose object is gone or that the federation mode restricts end without error. |
A local post, boost, follow, profile change, or other activity being recorded | Default. An unknown activity is dead at once. | Safe |
federation.deliver |
Signs and sends one delivery. Skips a suspended or unavailable server. A server behind an open circuit breaker counts as a failed attempt. | federation.fanout |
Up to 9 runs over about 4 days: 1 minute, 5 minutes, 30 minutes, 2 hours, 12 hours, then 24 hours. A missing delivery, activity, or signing key is dead at once. | Sends again. Inboxes ignore an activity they have seen. |
federation.update_poll |
Sends the current counts of a local poll to other servers. Votes that arrive while it waits share one job. | A vote in a local poll, after a short delay, or the end of the poll | Default | Safe |
federation.fetch_object |
Fetches one remote post and queues the fetches of its ancestors and replies, within the limits in its payload. A post that cannot be fetched or is not taken ends the job. | A post whose thread is incomplete | Default, only for failures of this server’s stores | Safe |
federation.fetch_outbox |
Fetches the first page of a remote account’s outbox and stores its recent public posts. At most one per account per day. | A local user viewing a remote account | Default, only for failures of this server’s stores | Safe |
federation.fetch_featured |
Fetches a remote account’s featured collection and makes its pins match. A collection that cannot be fetched leaves the pins as they are. | A remote actor being stored, or an activity about its pins or featured hashtags | Default, only for failures of this server’s stores | Safe |
federation.fetch_featured_tags |
Fetches a remote account’s featured hashtags collection and makes its featured hashtags match, up to 10. A collection that cannot be fetched leaves them as they are. | A remote actor being stored, or an activity about its pins or featured hashtags | Default, only for failures of this server’s stores | Safe |
federation.verify_quote |
Fetches the post a remote post quotes and verifies the quote’s approval. Without an approval the quote stays pending. | A remote post with a quote | Default, only for failures of this server’s stores | Safe |
federation.refetch_status |
Fetches one stored remote post again and records what its origin serves now: a revision, poll counts, or its deletion. | federation.refetch_walk, or my admin refetch status |
Default, only for failures of this server’s stores | Safe |
federation.refetch_walk |
Walks the remote posts of an account or domain, newest first, and queues a federation.refetch_status for each, one batch per minute. |
my admin refetch account or domain |
Default | Safe |
federation.refresh_actor |
Fetches one remote actor again. A fetch failure is recorded on the actor, and the periodic refresh tries again later. | federation.refresh_stale_actors |
Default, only for failures of this server’s stores | Safe |
federation.refresh_stale_actors |
Walks the stale remote actors in batches and refreshes those that local accounts follow. | Schedule | Default | Next run |
federation.observe_instance |
Reads the NodeInfo of a server seen for the first time. | A document from a new server being stored | Default. A server without NodeInfo ends the job. | Safe |
federation.probe_instances |
Fetches the NodeInfo of unavailable servers, 50 at a time, and marks those that answer as available again. | Schedule | Default | Next run |
federation.purge_records |
Deletes the inbox and delivery records past their retention: most after 30 days (the first delivery of a post stays until 30 days after the post or its author is deleted), and what is known about other servers after 90 days. | Schedule | Default | Next run |
Posts, media, and link cards
Section titled “Posts, media, and link cards”| Job type | What it does | Started by | On failure | Run again |
|---|---|---|---|---|
media.process |
Re-encodes one upload into its variants, asks the media.check hook about a new upload when it has a plugin, records the result, and deletes the received file. Content the processor rejects, and an upload the hook rejects, records a failed result. |
An upload, or a change of what a file is used for | Default | Safe |
media.purge_unattached |
Deletes the uploads that nothing attached in time. A file that a post or a draft attaches is attached, as are the cover and the body images of an article or of its draft; a file of a deleted draft goes with the next run. | Schedule | Default | Next run |
media.purge_replaced |
Deletes the old versions of images that a newer version replaced more than media_replaced_retention ago. An image’s current version is never deleted. |
Schedule | Default | Next run |
media.reprocess_walk |
Walks the uploaded images last processed before a time and queues a media.reprocess for each, one batch per minute. |
my admin media reprocess |
Default | Safe |
media.reprocess |
Processes one image again with the current settings, from its large version. An image processed since the time, or that the processor no longer accepts, stays as it is. | media.reprocess_walk |
Default | Safe |
media.purge_remote_cache |
Deletes the copies in the remote media cache past their expiry or over the size limit. | Schedule | Default | Next run |
content.close_poll |
Closes one poll when its end has passed. A job that runs early waits until the end. | A post with a poll, due at the poll’s end | Default | Safe |
content.publish_scheduled_draft |
Publishes one scheduled draft as a new post or article, with the checks of a new one at that time, and deletes the draft. A draft that cannot be published (the author may not post, the post it replies to or quotes is gone, a file is not usable, the checks or the automated review refuse it) stays a draft, its schedule ends as failed, and the author gets a scheduled_status_failed notification. A schedule that was cancelled or moved ends the job without doing anything. A job that runs early waits until the time, also beyond the delay limit of the queue. |
Scheduling a draft, due at its time | Default, only for failures of this server’s stores and the automated review that may pass | Safe: a draft is published once |
content.fetch_link_card |
Fetches the card of the URL a post links. A busy host, or a host that answers 429 or 503, postpones the job without counting as a failure. A robots.txt that cannot be read retries an hour later. Details are in Link previews. |
A post created or revised, after a delay for a remote post | Default, for failures of this server’s stores | Safe |
content.purge_link_cards |
Deletes the cards no post refers to and the images no card uses. | Schedule | Default | Next run |
translation.purge_cache |
Deletes the translations of posts 30 days after they were made, however often they were read, and the counts of translated characters from before last month. | Schedule | Default | Next run |
content.auto_delete_statuses |
Deletes local posts past their authors’ automatic deletion policies, as their authors would, at most 100 per account and run. While the server sets a maximum for keeping posts, reads every local account: tells each account that the maximum newly applies to, and holds it to the maximum once its grace period ends. | Schedule, then itself for the next batch | Default | Next run. Each account is told once per maximum. |
content.purge_remote_statuses |
Deletes remote posts past their retention, in batches, and after 30 days the records of those deletions that no interaction or notification still needs. | Schedule, then itself | Default | Next run |
content.purge_relayed_statuses |
Deletes the posts that only relays brought in past their retention, in batches. | Schedule, then itself | Default | Next run |
content.purge_deleted_statuses |
Deletes the text, media references, and polls of the posts deleted more than 7 days ago, in batches; the record of each deleted post’s URI stays. A post that a report or an action points to, or whose author is under a hold, stays. | Schedule, then itself | Default | Next run |
content.purge_remote_emojis |
Deletes the remote custom emoji nothing refers to, in batches. | Schedule, then itself | Default | Next run |
content.purge_boards |
Deletes the posts of the boards of events that ended or were cancelled past event_records_retention_days, except those of an event a report points to, in batches; their text and media go with content.purge_deleted_statuses. |
Schedule | Default | Next run |
Accounts, follows, and imports
Section titled “Accounts, follows, and imports”| Job type | What it does | Started by | On failure | Run again |
|---|---|---|---|---|
account.execute_deletions |
Deletes the accounts whose scheduled deletion is due, from a ban or from the owner’s request, and the accounts that their owners deactivated 30 days ago. An account that came back, or whose deletion was cancelled meanwhile, is left alone, and so is a deactivated account that is the last administrator, with a warning in the log. | Schedule | Default | Next run |
account.erase_deleted |
Erases the data of deleted accounts, local and remote, in bounded batches. First fetches a remote account again when only a 410 Gone marked it deleted. |
Schedule, then itself | Default | Next run |
account.prune_remote_profiles |
Shrinks the profile history of remote accounts that nothing local relates to, in batches, and deletes checks of remote handles after 90 days, keeping the newest of each account. | Schedule, then itself | Default | Next run |
graph.expire_mutes |
Deletes the mutes that have expired. | Schedule | Default | Next run |
graph.migrate_follows |
Moves one page of the relationships of a moved account, waiting 10 seconds between pages to spread its deliveries. | An account move being recorded | Default. A step whose data is gone is dead at once. | Safe. It does nothing once the move is cancelled or replaced. While the move is in force, a run much later also moves the accounts that follow, block, or mute the account by then. |
identity.process_import |
Carries out one batch of a user’s data import and records each item’s result. An item that keeps failing is recorded as failed after 5 tries, and the import goes on. | A user’s import, then itself | Default | Safe |
identity.expire_imports |
Removes the imports past their lifetime. | Schedule | Default | Next run |
identity.build_export |
Builds the archive of a user’s data export. | A user’s export request | Up to 8 runs: 5 that build, 3 that only record the failure, with the default waits. A failed export is recorded for the user, not left as a dead job. | Safe. An export with a result stays as it is. |
identity.expire_exports |
Removes expired archives and records, and records as failed an export still without a result after 24 hours. | Schedule | Default | Next run |
identity.purge_unused_oauth_clients |
Deletes the dynamically registered OAuth clients that nothing uses. | Schedule | Default | Next run |
identity.purge_origins |
Deletes the session and registration origins past their retention. | Schedule | Default | Next run |
identity.purge_auth_records |
Deletes the sessions, authorization codes, token families, and one-time tokens that expired over 30 days ago. A token family goes only when its last token did. | Schedule | Default | Next run |
Mail jobs run when mail is set to a provider. With mail = none, only identity.verification_email and identity.password_notice_email run: they send nothing and log the category of the message, or, with profile = development, the whole message. If the address or the state they check has changed, they end without sending.
| Job type | What it does | Started by | On failure | Run again |
|---|---|---|---|---|
identity.verification_email |
Issues a token and sends the verification link. Sends nothing more than an hour after the request. | Registration, or a request to send the link again | Transactional mail | Sends again, within the hour, with a new link |
identity.password_reset_email |
Issues a token and sends a password reset link. Sends nothing more than 15 minutes after the request, or while sign-in is limited to login providers. | A password reset request | Transactional mail | Sends again, within the 15 minutes, with a new link |
identity.password_notice_email |
Tells a user that their password changed or was removed. | A password change or removal | Sends again | |
identity.deactivation_email |
Tells a user that their account is deactivated, with the date of its deletion. Sends nothing once the user has come back. | A user deactivating their account | Sends again | |
notification.email |
Sends the email of one notification that stayed unread for 15 minutes. Checks the read position, the user’s setting, and that the notification is still shown. | A notification with email on, due 15 minutes later | Sends again | |
notification.moderator_email |
Sends a moderator notification by email, without waiting. | A moderator notification | Sends again | |
notification.supporter_email |
Sends a supporter notification by email, unless the subscriber has deactivated their account since. | A subscription’s transition, or a plan being retired | Sends again | |
identity.closure_email_fanout |
Queues one service closure email for every user with a verified primary address, page by page, each user once. Above 1,000 of them, the first job only queues jobs of the same type for up to 1,000 each. | The closure being recorded, or identity.queue_closure_reminders |
Safe | |
identity.closure_email |
Sends one service closure email to one user. | identity.closure_email_fanout |
Sends again | |
identity.queue_closure_reminders |
Queues the fan-out of each closure reminder whose time has come. | Schedule | Default | Next run |
Notifications
Section titled “Notifications”| Job type | What it does | Started by | On failure | Run again |
|---|---|---|---|---|
notification.push |
Sends one notification to one Web Push subscription, if the subscription is still valid, the notification is still shown, and the recipient has not deactivated their account since. A subscription the push service reports gone, or whose sign-in has ended, is removed. | A notification, for each subscription that wants its type | Default, using the wait the push service asks for. A refusal, or a message that cannot be sent, is dead at once. | Sends again. A device shows a repeat in place of the first message. |
notification.post_subscribers |
Notifies the subscribers of a post’s author. Above 1,000 of them, the first job only queues jobs of the same type for up to 1,000 each. | A new local post, or a delivered remote post; a reply that waited for its approval, once it is approved | Default | Safe. Each notification is created once. |
instance.notify_announcement |
Creates the announcement notifications for local users. Waits if the announcement has not started, and ends if it has ended. Above 1,000 of them, the first job only queues jobs of the same type for up to 1,000 each. | An announcement getting the time it starts to show | Default | Safe. Each notification is created once. |
notification.closure_fanout |
Creates the service closure notification for every local user. Above 1,000 of them, the first job only queues jobs of the same type for up to 1,000 each. | The closure being recorded | Default | Safe. Each notification is created once. |
notification.closure_due |
Creates the notification for administrators once the closing date has passed and the announcement is still in effect. | The closure being recorded, due at the closing date | Default | Safe. The notification is created once. |
notification.purge_error_group_notices |
Deletes the hourly error notices past their retention. | Schedule | Default | Next run |
Moderation and service closure
Section titled “Moderation and service closure”| Job type | What it does | Started by | On failure | Run again |
|---|---|---|---|---|
moderation.sever_domain |
Cuts the follows of local accounts with a suspended domain, and records what each account lost. Above 1,000 of them, the first job only queues jobs of the same type for up to 1,000 each. Running one again repeats nothing it did. | A domain suspension | Default. An action that is gone or is not a domain suspension is dead at once. | Safe |
moderation.sever_allowlist |
Cuts the follows of local accounts with the servers that the allowlist of federation cuts off (when federation_scope becomes allowlist, or an allow is removed while it is), and records what each account lost. It cuts nothing once federation_scope is open again or the server is allowed again. Above 1,000 local accounts, the first job only queues jobs of the same type for up to 1,000 each. Running one again repeats nothing it did. |
The switch to allowlist, or the removal of an allow |
Default. A cutoff that is gone is dead at once. | Safe |
moderation.expire_restrictions |
Removes expired account and domain restrictions, tells other servers when a suspension ends, and with a search backend queues search.sync_author or search.sync_domain for what they restricted. |
Schedule | Default | Next run |
moderation.purge_records |
Erases the moderation records past their retention, including released retention holds. | Schedule | Default | Next run |
moderation.fetch_domain_lists |
Queues moderation.fetch_domain_list for every subscribed domain block list. |
Schedule | Default | Next run |
moderation.fetch_domain_list |
Fetches one subscribed domain block list, records the fetch, and replaces the domains the subscription holds. In the automatic mode it then issues and lifts the subscription’s domain actions. A failed fetch is recorded and changes nothing else. | moderation.fetch_domain_lists, a new subscription or new settings of one, and Fetch now |
Default. A subscription deleted since is done at once. | Safe |
identity.reject_for_closure |
Rejects the pending registrations when the closure’s notice period starts. Stops if the closure is withdrawn. | The closure being recorded | Default | Safe |
instance.closure_delete_actors |
Records the Delete of one page of local accounts and passes them to the fan-out. |
The closure being recorded, then itself | Default | Safe |
instance.check_closure_delivery |
Records that the closure’s deletes have all been delivered. | Schedule | Default | Next run |
Supporters and costs
Section titled “Supporters and costs”| Job type | What it does | Started by | On failure | Run again |
|---|---|---|---|---|
supporters.ensure_prices |
Makes the prices of a plan’s current terms at the payment provider, and archives those no longer used. | A plan’s terms changing | Default | Safe. The provider call is idempotent. |
supporters.sync |
Reads a subscription again from the provider and applies what it finds. | Provider notices, a due deadline, a checkout | Default | Safe. It only re-reads and applies. |
supporters.command |
Sends a cancellation to the provider, removes what a failed plan change left to collect, expires the payment pages of a closed pending subscription, or cancels and refunds a second subscription of one payment. | A user’s or the server’s cancellation, a pending subscription closing, a second subscription found, each reconciliation while one is not settled | Default | Safe. A repeat finds the page expired, the subscription ended, and the charge refunded. |
supporters.sync_due |
Queues supporters.sync for the subscriptions whose deadline has passed, and supporters.sync_tip for the open tips whose page has ended. |
Schedule | Default | Next run |
supporters.sync_tip |
Reads the page of an open tip again from the provider, or makes it unpayable first when the account closed, and records its payment or its end. | Reading an open tip, a tip’s page ending, an account’s deletion | Default | Safe. It only re-reads and applies, and a payment is recorded once. |
supporters.cancel_for_closure |
Cancels the subscriptions when a service closure is announced, and tells the subscribers. | The closure being recorded | Default | Safe |
supporters.apply_payment_event |
Applies one notice from the payment provider by re-reading what it points at. | A verified webhook from the provider | Default | Safe. Duplicates and reordering come down to re-reading. |
supporters.reconcile_payments |
Compares the provider’s records with this server’s, within a request limit, and continues in a later job when it uses them up. A provider failure is recorded and waits for the next run. | Schedule, then itself | Default, for failures of this server | Next run |
supporters.purge_payment_receipts |
Deletes old webhook receipts and expired checkout claims. | Schedule | Default | Next run |
finance.import_billing |
Imports the charges of one source of billing_sources. A run that already succeeded for its key does nothing, and a source over its request limit ends without retrying. See Importing hosting costs. |
Schedule, or an administrator, at most once an hour | Default, for throttling or an unavailable source | Next run |
| Job type | What it does | Started by | On failure | Run again |
|---|---|---|---|---|
voice.apply |
Applies one change of a room to its media transport: open, allow or stop audio, mute, remove, or close. Changes of one room run in order. | A committed change of a room | Default. Once it is dead, voice.sweep reconciles the room. |
Safe. Opening does nothing for a room that ended, and allowing or stopping audio applies the participant’s role now. A mute mutes the session again, and a removal or close that is done stays done. |
voice.apply_media_event |
Applies one verified event from a media server to its room. | A webhook from a media server | Default | Safe. It compares the event with the room and the media server as they are now. |
voice.sweep |
Evaluates every open room: participants’ expiry, hosts’ grace, the wait of rooms opened for a scheduled time, rooms of a voice chat turned off, and rooms over the time limit. | Schedule | Default | Next run |
voice.check_account |
Applies a change of an account to its participation: a suspended or deleted account leaves, and a host who is no longer eligible hands over. | A suspension, deletion, or restriction of an account in a room | Default | Safe. It reads the account and the room as they are now. |
voice.check_block |
Applies a block between two participants of a room: the one who is not a host or co-host leaves. | A block | Default | Safe. A block that was lifted makes nobody leave. |
voice.check_hosts |
Checks the eligibility of every open room’s host. | A change of roles or of voice_room_starters while rooms are open |
Default | Safe. It reads the hosts and the roles as they are now. |
voice.invite_scheduled |
Copies up to 500 invited people into a room opened for a scheduled time, as listeners invited by its host, without notifying them. | A room opening for a scheduled time, or more people added to it | Default | Safe. Each person is invited to a room once. |
voice.notify_started |
Tells the local followers of a host who can see a room that it opened. Above 1,000 of them, the first job only queues jobs of the same type for up to 1,000 each. | A room opening | Default | Safe. Each notification is created once. |
voice.purge_records |
Erases the participants’ records of rooms that ended past their retention, except those a report or a hold preserves. | Schedule | Default | Next run |
Events
Section titled “Events”| Job type | What it does | Started by | On failure | Run again |
|---|---|---|---|---|
gathering.tick |
Reads the events whose next reminder, check of the voice chat room one day before the start, or start has come, queues gathering.notify for each reminder and start, and gathering.room for each check and for the start of an event with a voice chat room, that came since, and sets the time of the next one. Handles up to 1,000 events a run; the rest wait for the next run. |
Schedule | Default | Next run |
gathering.room |
At the start of an event with a voice chat room, opens the room for its date, hosted by the organizer, and invites the co-organizers, those who answered going or maybe, and the invited who have not answered, without notifying them. A room that cannot open is recorded, and the organizer is notified. When the date is set or changed and one day before the start, it only checks that the organizer can open the room, and notifies the organizer when not. Does nothing for a date that changed since, or an event cancelled since or without a voice chat room. | A date set or changed; gathering.tick |
Default | Safe. A date opens one room, and each notification is created once. |
gathering.end_rooms |
Ends the voice chat rooms of an event that was cancelled or whose voice chat room was removed, as their host, or as a moderator for a moderator’s cancellation. The event stays. | A cancellation, or a change that removes the voice chat room; gathering.room when the event was cancelled while its room opened |
Default | Safe. A room that ended stays as it ended. |
gathering.notify |
Sends one notification of an event: an invitation, the date confirmed from its candidate dates, a change of the date or place, a cancellation, a reminder, the start, a voice chat room that cannot open, to its organizer, or a new post on its board, to those who read the board and answered going or maybe. It picks the recipients when it runs, and only those who can still see the event (or read its board). Above 500 of them, the first job only queues jobs of the same type for up to 500 each. A reminder or start of a date that changed since, or of an event cancelled since, sends nothing. | An invitation, a confirmed date, a change, a cancellation, or a board post; gathering.tick and gathering.room; a voice chat room of an event that nobody came to |
Default | Safe. Each notification is created once. |
gathering.cancel_organized |
Cancels the events of a suspended or deleted organizer that have not started, date polls not yet confirmed included, each as a cancellation that notifies its attendees. Cancels nothing when the account is active again by the time it runs. | A suspension or deletion of an account that organizes an upcoming event | Default | Safe. A cancelled event is not cancelled again. |
gathering.purge_records |
Erases the answers, approvals, invitations, and votes on candidate dates of the events that ended or were cancelled past event_records_retention_days, except those a report or a hold preserves. |
Schedule | Default | Next run |
Full-text search
Section titled “Full-text search”These types keep the index of a search backend in step with the database. Each job carries an ID and reads the current state when it runs, so repeating or reordering them gives the same index. Every type exists on every server, and with search = none, the default, nothing enqueues them and any that remain succeed doing nothing. Each call of the backend has a 10-second limit, outside any database transaction.
| Job type | What it does | Started by | On failure | Run again |
|---|---|---|---|---|
search.sync_status |
Writes the document of one post if it qualifies and removes it if not, then reads the post again and repeats if it stopped qualifying or was edited meanwhile. | A post created, edited, or deleted, local or remote, when its visibility is public or local |
Default | Safe. It reads the post as it is now. |
search.sync_author |
Removes every document of an account that cannot be found, and otherwise syncs its posts. Above 200 posts, the first job only queues jobs of the same type for up to 200 each. | An account turning indexable on or off, a lifecycle event, a sanction, its lift, its expiry, a job of search.sync_domain, or the erasure of a deleted account |
Default | Safe. It reads the account and its posts as they are now. |
search.sync_domain |
Queues search.sync_author for each account of a domain and its subdomains. Above 1,000 of them, the first job only queues jobs of the same type for up to 1,000 each. |
A domain action, its lift, or its expiry | Default | Safe. The jobs it queues read the accounts as they are now. |
search.narrow |
Removes every document of a visibility, unless search_visibilities includes it again. |
A change of search_visibilities that leaves public or local out |
Default | Safe. It reads the setting as it is now. |
search.rebuild |
Handles the next 100 public and local posts of a rebuild of the index the same way as search.sync_status, records its progress, and queues the next job at the rate of the rebuild. |
my search rebuild, and the job before it |
Default | Safe. A job that finds the rebuild already past its place does nothing. |
Webhooks
Section titled “Webhooks”| Job type | What it does | Started by | On failure | Run again |
|---|---|---|---|---|
webhook.deliver |
Sends one delivery to an endpoint. Sends nothing if the endpoint is not enabled or the delivery already succeeded, unless it is a redelivery. | An event an endpoint subscribes to, or a redelivery | Up to 8 runs over about 27 hours. See the Webhook reference. | Sends again, only if it has not succeeded |
webhook.purge_messages |
Deletes messages past their retention, with their deliveries and attempts. | Schedule | Default | Next run |
Upkeep
Section titled “Upkeep”| Job type | What it does | Started by | On failure | Run again |
|---|---|---|---|---|
queue.purge_dead_jobs |
Deletes dead jobs after dead_job_retention and attempt records after job_attempt_retention. |
Schedule | Default | Next run |
kv.purge_expired |
Deletes the expired entries of the key-value store. | Schedule | Default | Next run |
metrics.snapshot |
Computes the gauges of the instance, stores them, and exports them once. | Schedule | Default | Next run |
projection.verify |
Rebuilds every projection, writing only what drifted, and logs those that did. One that fails does not stop the others. | Schedule | Default | Next run |
diagnostics.purge_error_counts |
Deletes the hourly error counts past their retention. | Schedule | Default | Next run |
diagnostics.purge_ux_counts |
Deletes the hourly counts of the UX measurements older than 30 days. | Schedule | Default | Next run |
hook.purge_decisions |
Deletes the decisions of the decision hooks past hook_decisions_retention_days, with their appeals, except those it keeps. |
Schedule | Default | Next run |
stream.sweep |
Deletes the per-process resources that ended processes left on the broker. | Schedule | Default | Next run |
Periodic jobs
Section titled “Periodic jobs”The scheduler queues one job per period, however many processes run it, and the job then runs like any other. With queue = cloudtasks the platform’s scheduler starts the periods; see
The job queue and failed (dead) jobs.
| Periodic job | Every |
|---|---|
gathering.tick |
1 minute |
voice.sweep |
1 minute |
moderation.expire_restrictions |
5 minutes |
supporters.sync_due |
5 minutes |
instance.check_closure_delivery |
10 minutes |
metrics.snapshot |
15 minutes |
stream.sweep |
15 minutes |
account.erase_deleted |
1 hour |
account.execute_deletions |
1 hour |
content.auto_delete_statuses |
1 hour |
content.purge_deleted_statuses |
1 hour |
content.purge_boards |
1 hour |
content.purge_relayed_statuses |
1 hour |
content.purge_remote_statuses |
1 hour |
federation.purge_records |
1 hour |
federation.refresh_stale_actors |
1 hour |
gathering.purge_records |
1 hour |
graph.expire_mutes |
1 hour |
identity.queue_closure_reminders |
1 hour |
identity.purge_unused_oauth_clients |
1 hour |
identity.expire_exports |
1 hour |
identity.expire_imports |
1 hour |
identity.purge_auth_records |
1 hour |
identity.purge_origins |
1 hour |
kv.purge_expired |
1 hour |
media.purge_unattached |
1 hour |
media.purge_replaced |
1 hour |
media.purge_remote_cache |
1 hour |
moderation.purge_records |
1 hour |
queue.purge_dead_jobs |
1 hour |
voice.purge_records |
1 hour |
webhook.purge_messages |
1 hour |
account.prune_remote_profiles |
24 hours |
content.purge_link_cards |
24 hours |
moderation.fetch_domain_lists |
24 hours |
content.purge_remote_emojis |
24 hours |
diagnostics.purge_error_counts |
24 hours |
diagnostics.purge_ux_counts |
24 hours |
hook.purge_decisions |
1 hour |
federation.probe_instances |
24 hours |
finance.import_billing |
24 hours, one schedule per source of billing_sources, named finance.import_billing.<source> |
notification.purge_error_group_notices |
24 hours |
translation.purge_cache |
24 hours |
projection.verify |
24 hours |
supporters.purge_payment_receipts |
24 hours |
supporters.reconcile_payments |
payments_reconcile_interval (6 hours by default) |
Jobs that depend on settings
Section titled “Jobs that depend on settings”Most job types exist on every server. These do not:
| Setting | Job types |
|---|---|
mail set to a provider (not none) |
notification.email, notification.moderator_email, notification.supporter_email |
vapid_private_key set |
notification.push |
stream = sns or pubsub |
stream.sweep |
payments set |
supporters.apply_payment_event, supporters.reconcile_payments, supporters.purge_payment_receipts |
billing_sources set |
The finance.import_billing schedules. The job type itself always exists. |
A job whose type has no handler on the process that claims it, for example after you change one of these settings, counts as a failure, retries with the default waits, and becomes a dead job. It appears as other in the my.job.* metrics. Restore the setting and run it again, or delete it.
These types always exist but do nothing without their setting: supporters.ensure_prices, supporters.sync, and supporters.command without payments, voice.apply without a voice transport, and the search.* types with search = none (Full-text search). The other voice.* jobs run with the voice chat off, so that voice.sweep can end rooms left open.
Para LLMs e agentes de programação:llms.txt,llms-full.txt, llms-small.txt