async-io-model
Explanations of common asynchronous patterns used in tursodb. Involves IOResult, state machines, re-entrancy pitfalls, CompletionGroup. Always use these…
Use when working with Turso yield injection or synthetic failure injection machinery for safe resumable state-machine boundaries, including YieldInjector, FailureInjector, YieldPointMarker, inject_transition_yield!, inject_io_yield!, inject_transition_failure!, CommitYieldPoint,
$ npx -y skills add tursodatabase/turso --skill yield-injections --agent claude-codeHow it fires
How this skill gets triggered: by you, by Claude, or both.
/yield-injectionsContext preview
The summary Claude sees to decide when to auto-load this skill.
Use when working with Turso yield injection or synthetic failure injection machinery for safe resumable state-machine boundaries, including YieldInjector, FailureInjector, YieldPointMarker, inject_transition_yield!, inject_io_yield!, inject_transition_failure!, CommitYieldPoint,
name: yield-injections description: Use when working with Turso yield injection or synthetic failure injection machinery for safe resumable state-machine boundaries, including YieldInjector, FailureInjector, YieldPointMarker, inject_transition_yield!, inject_io_yield!, inject_transition_failure!, CommitYieldPoint, CheckpointYieldPoint, CursorYieldPoint, deterministic interleaving tests, abandoned statement/commit tests, and concurrent-simulator yield plans.
Yield injection is test-only infrastructure for forcing cooperative-yield or failure boundaries in resumable state machines. The implementation currently lives under `core/mvcc/`, but the mechanism is not MVCC semantics.
All hooks are behind `cfg(any(test, injected_yields))`; `injected_yields` means `test_helper` or `simulator`.
`YieldPoint { ordinal, point_count }` identifies one hook in one point family. Families are `#[repr(u8)]` enums implementing `YieldPointMarker`; `ordinal` is source-order `self as u8`, and `point_count` comes from `EnumCount`.
Do not reorder existing yield-point variants. Simulator plans store raw ordinals, not names.
Why reordering matters: `ordinal` is `self as u8`, so source order is part of the deterministic schedule. If a seed planned "yield at ordinal 2", reordering can silently retarget that same seed from one conceptual hook to another. This makes CI seeds, bisects, and simulator coverage hard to reproduce. Appending a variant still changes `point_count` and can perturb future plans, but it preserves the meaning of existing ordinals.
Each yield-capable live object has:
There is no `inject_io_failure!` today; failure injection only works for `TransitionResult` state machines.
Injected yields return `Completion::new_yield()`. It is already finished, but `is_explicit_yield()` is true, so the VDBE surfaces it as `StepResult::Yield` instead of continuing immediately.
For a new state machine/cursor family, add all of this behind `cfg(any(test, injected_yields))`:
yield_instance_id: u64,
Initialize it from the connection:
yield_instance_id: connection.next_yield_instance_id(),
Implement `ProvidesYieldContext`:
impl ProvidesYieldContext for MyStateMachine {
fn yield_context(&self) -> YieldContext {
YieldContext::new(
self.connection.yield_injector(),
self.connection.failure_injector(),
self.yield_instance_id,
my_yield_key(self.logical_operation_id),
)
}
}Add a family-specific `*_yield_key(...) -> u64` helper. Mix stable logical identity, such as tx id/table id, with a family tag so simulator plans do not collide across families. Do not use wall-clock time, random values, allocation addresses, or incidental counters unrelated to the logical operation.
Before adding a hook, prove re-entry is safe. A synthetic yield returns `StepResult::Yield`; the same statement/state machine may be stepped again immediately later. On re-entry, it must resume from an explicit state, not repeat non-idempotent work.
Rules:
Checklist:
1. Append a variant to the appropriate `*YieldPoint` enum. Do not reorder. 2. Place the macro after the transition that makes resume safe, or before a lock acquisition when explicitly testing lock interleavings. 3. Avoid calling macros while borrowing `&mut self.state`; they need `self.yield_context()`. 4. Add a deterministic test with `FixedYieldInjector` or `FixedFailureInjector`. 5. If testing abandoned work, drop the statement at the yield and assert cleanup invariants.
For deeper re-entry/state-machine rules, also use `async-io-model`.
Use fixed injectors in targeted tests:
use crate::mvcc::yield_hooks::YieldPointMarker;
conn.set_yield_injector(Some(FixedYieldInjector::new([
CommitYieldPoint::LogRecordPrepared.point(),
])));`FixedYieldInjector` stores a `HashSet<YieldPoint>`, ignores `instance_id`/`selection_key`, and consumes each configured point once total. If two simultaneous instances hit the same point, th
A SQL database in Rust: SQLite-compatible, now also speaking Postgres (experimental). The LLVM of databases.
Repo: tursodatabase/turso
Explanations of common asynchronous patterns used in tursodb. Involves IOResult, state machines, re-entrancy pitfalls, CompletionGroup. Always use these…
Change Data Capture - architecture, entrypoints, bytecode emission, sync engine integration, tests
General Correctness rules, Rust patterns, comments, avoiding over-engineering. When writing code always take these into account
How to debug tursodb using Bytecode comparison, logging, ThreadSanitizer, deterministic simulation, and corruption analysis tools
Information about the differential fuzzer tool, how to run it and use it catch bugs in Turso. Always load this skill when running this tool
Generate hierarchical AGENTS.md knowledge base for a codebase. Creates root + complexity-scored subdirectory documentation.