Instrucciones de origen · Vista previa de solo lectura
name
add-background-task
description
Add a new Nexus background task. Use when the user wants to create a periodic background task in Nexus that runs on a timer.
Add a Nexus background task
All background tasks live in Nexus. A task implements the BackgroundTask trait (nexus/src/app/background/mod.rs), runs on a configurable period, and reports status as serde_json::Value.
General approach
There are many existing background tasks in nexus/src/app/background/tasks/. Before writing anything, read a few tasks that are similar in shape to the one you're adding (e.g., a simple periodic cleanup vs. a task that watches a channel). Use those as models for structure, naming, logging, error handling, and status reporting. The goal is to conform to the patterns already in use, not to invent new ones.
Checklist
These are the touch points for adding a new background task. Follow them in order.
1. Status type (nexus/types/src/internal_api/background.rs)
Define a struct for the task's activation status. Derive Clone, Debug, Deserialize, Serialize, PartialEq, Eq. For errors, use Option<String> if the task can only fail in one way per activation, or Vec<String> if it accumulates multiple independent errors. Match what similar tasks do.
Create the task module. The struct holds whatever state it needs (typically Arc<DataStore> plus config). Implement BackgroundTask::activate by delegating to an actually_activate helper, then serialize the status to serde_json::Value. The actually_activate pattern makes unit testing easy without going through the trait.
actually_activate can either build and return the status (), or take a mutable reference to one (). The first is simpler and works well when the task either fully succeeds or fully fails. The second is better when the task can partially complete (e.g., it loops over work items): creates the status struct up front, passes it in, and serializes it afterward regardless of /, so any progress already recorded in (items processed, partial counts, earlier errors) is preserved even if the method bails out with later.
Logging conventions: debug when there's nothing to do, info when routine work was done, warn when the work done indicates something is wrong (e.g., cleaning up after a crash), error on failure. Log errors as structured fields with the ; &err slog syntax (which uses the SlogInlineError trait), not by interpolating into the message string. For the error string in the status struct, use InlineErrorChain::new(&err).to_string() (from slog_error_chain) to capture the full cause chain. Status error strings should not repeat the task name — omdb already shows which task you're looking at.
If the task takes config values that need conversion or validation (e.g., converting a Duration to TimeDelta, or checking a numeric range), do it once in new() and store the validated form. Don't re-validate on every activation — if the config is invalid, panic in new() with a message that includes the invalid value.
Include a unit test in the same file using TestDatabase::new_with_datastore that calls actually_activate directly. If the task has a datastore method, a single test exercising the task end-to-end (including the limit/batching behavior) is sufficient — don't add a redundant test for the datastore method separately unless it has complex logic worth testing in isolation.
For tasks that mutate or delete a bounded set of rows, the test should prove which items were affected, not just how many. Seed a mix of rows that should be affected, rows that should be skipped because they're on the safe side of the cutoff or predicate, and rows that should be skipped because they're structurally ineligible. Assert exact identities of the affected and surviving rows after each activation. If the task uses ordering plus a per-activation limit, make the first activation hit the limit and verify that the highest-priority eligible items were chosen, then run a second activation to verify the remainder, and a final activation to verify the no-op case. When eligibility depends on time, set timestamps explicitly in the test rather than relying on sleeps. When selection depends on multiple sort keys, include a case that exercises the tiebreaker. Prefer assertions on row IDs or full records over aggregate counts and status fields.
3. Register the module (nexus/src/app/background/tasks/mod.rs)
Add pub task_<name>: Activator to the BackgroundTasks struct, maintaining alphabetical order among the task fields.
5. Config (nexus-config/src/nexus_config.rs)
Add a config struct (e.g., YourTaskConfig) with at minimum period_secs: Duration (using #[serde_as(as = "DurationSeconds<u64>")]). If the task does bounded work per activation, name the limit field max_<past_tense_verb>_per_activation (e.g., max_deleted_per_activation, max_timed_out_per_activation) to match existing conventions. Add the field to BackgroundTaskConfig. Update the test config literal and expected parse output at the bottom of the file.
6. Config files
Add the new config fields to all of these:
nexus/examples/config.toml
nexus/examples/config-second.toml
nexus/tests/config.test.toml
smf/nexus/single-sled/config-partial.toml
smf/nexus/multi-sled/config-partial.toml
7. Wire up in nexus/src/app/background/init.rs
Import the task module.
Add Activator::new() in the BackgroundTasks constructor.
Destructure it in the start method.
Call driver.register(TaskDefinition { ... }) with the task. The last task registered should pass datastore by move (not .clone()), so adjust the previous last task if needed.
If extra data is needed from BackgroundTasksData, add the field there and plumb it from nexus/src/app/mod.rs.
8. Schema migration (if needed)
If the task needs a new index or schema change to support its query, add a migration under schema/crdb/. See schema/crdb/README.adoc for the procedure. Also update dbinit.sql and bump the version in nexus/db-model/src/schema_versions.rs.
9. Datastore method (if needed)
If the task needs a new query, add it in the appropriate nexus/db-queries/src/db/datastore/ file. Prefer the Diesel typed DSL over raw SQL (diesel::sql_query) for queries and test helpers. Only fall back to raw SQL when the DSL genuinely can't express the query.
If the task modifies rows that other code paths also modify, think about races: what happens if both run concurrently on the same row? Both paths should typically guard their writes so only one succeeds.
Add a print_task_<name> function and wire it into the match in print_task_details (alphabetical order). Import the status type. Use the const_max_len + WIDTH pattern to align columns:
