autowisp.tests.test_error_context module

Class Inheritance Diagram

Unit tests for the ambient error context and capture layer.

The in-process tests need no pipeline fixtures or database; the cross-process tests spin up a real multiprocessing.Pool (through run_pool) and let its worker bootstrap with the real setup_process_map, which initialises a throwaway project home in a temporary directory. Each test starts from a clean ambient context (reset in setUp).

class autowisp.tests.test_error_context.TestAmbientAccessors(methodName='runTest')

Bases: _ContextTestCase

get/set helpers and in_worker().

test_set_and_get()

set_error_context installs a bundle get_error_context returns.

test_set_pipeline_run_keeps_rest()

set_pipeline_run swaps only the run, preserving other fields.

class autowisp.tests.test_error_context.TestCaptureErrors(methodName='runTest')

Bases: _ContextTestCase

The capture_errors decorator stamps + optionally wraps.

test_bare_value_error_wrapped_to_step_subclass()

A bare ValueError becomes the ambient step’s StepError type.

test_existing_autowisp_error_passes_through_stamped()

An AutoWISPError keeps its type and gets stamped from context.

test_no_run_leaves_pipeline_run_none()

With no run in context, pipeline_run/crashed stay unset.

test_pipeline_and_bui_wrapping()

Component selects the wrapper class.

test_preset_step_name_not_overwritten()

A step name set at the raise site survives stamping.

test_success_returns_value()

The happy path returns the wrapped callable’s result.

test_unknown_step_falls_back_to_base_steperror()

An unmapped step name yields a plain StepError.

test_wrap_unknown_false_passes_through()

With wrap_unknown=False the original exception is untouched.

class autowisp.tests.test_error_context.TestErrorContextDataclass(methodName='runTest')

Bases: _ContextTestCase

The ErrorContext bundle itself.

test_defaults()

A bare ErrorContext has empty/None fields.

test_frozen()

ErrorContext is immutable.

class autowisp.tests.test_error_context.TestErrorContextManager(methodName='runTest')

Bases: _ContextTestCase

The error_context() scoping context manager.

test_keeps_pipeline_run_and_worker_flag()

The block inherits run + in_worker from the current context.

test_nesting_innermost_step_wins_and_files_accumulate()

Inner step overrides; related files from both levels stack.

test_scopes_and_restores()

Step + files apply inside the block and are removed after.

class autowisp.tests.test_error_context.TestExitSignalDecode(methodName='runTest')

Bases: TestCase

decode_exit_signals is portable and drops clean/running exits.

test_drops_running_and_clean()

None (running) and 0 (clean) contribute nothing.

test_posix_signals_and_plain_codes()

POSIX: negative codes decode to their signal; positive don’t.

test_windows_status_not_signal()

Windows: an abnormal code is an NTSTATUS, never a POSIX signal.

Both the unsigned and signed spellings of an access violation map to the conventional hex status; a plain small code stays bare.

class autowisp.tests.test_error_context.TestFromConfig(methodName='runTest')

Bases: _ContextTestCase

ErrorContext.from_config rebuilds context from a config dict.

test_code_version_fallback()

A missing code_version falls back to get_code_version_str().

test_empty_config()

No run keys -> no snapshot, no step, main process.

test_full_config()

Run keys produce a populated pipeline-run snapshot.

test_in_worker_from_parent_pid()

in_worker is True iff a truthy parent_pid is present.

test_sentinel_step_names_become_none()

Bootstrap sentinels are not treated as real step names.

class autowisp.tests.test_error_context.TestNestingGuard(methodName='runTest')

Bases: _ContextTestCase

A worker may not launch nested workers (resource control).

test_forbid_nested_workers_only_inside_worker()

The guard is a no-op in the main process, raises in a worker.

test_nested_run_pool_blocked_end_to_end()

A real worker that calls run_pool is refused, not allowed to nest.

Goes through the full chain: the outer run_pool spawns a real worker via setup_process_map (which sets in_worker), the worker calls run_pool again, and the inner guard raises a PipelineError that propagates back stamped with the worker PID – so an accidental N^2 launch fails loudly instead of spawning.

class autowisp.tests.test_error_context.TestPoolPropagation(methodName='runTest')

Bases: _ContextTestCase

Errors raised in real Pool workers cross back stamped.

test_bare_exception_wrapped_to_mapped_step_error()

A worker’s bare exception returns as the step’s StepError subclass.

The worker raises a plain ValueError; _wrap maps the ambient step name to its concrete subclass, so the parent receives a MeasurePhotometryError. The original is not kept as a live __cause__ across the Pool (our pickling carries only the fields, and multiprocessing replaces __cause__ with its RemoteTraceback); the durable record is the formatted traceback in details.

test_faulthandler_dumps_native_traceback()

A segfaulting worker leaves a native dump in its own log.

Proves the faulthandler armed in setup_process_map turns an otherwise-silent SIGSEGV into a collectable C-level traceback – which is what distinguishes the faulted worker from the innocents the executor merely terminates.

test_step_error_propagates_with_context()

A worker StepError returns same-typed, stamped, with traceback.

Exercises the full run_pool round trip: the worker bootstraps its context from config (via the real setup_process_map), raises a FindStarsError, and the parent receives the same type carrying the worker’s subprocess_id (a different PID), the step name and pipeline-run snapshot rebuilt in the worker, and the worker traceback both in details and as the Pool’s RemoteTraceback cause.

test_worker_crashed_carries_scoped_config()

A crash carries the failing step’s config, scoped by the manager.

The parent’s own context holds the base add_images_to_db config, so the failing step’s config must come from the error_context(config=...) the manager scopes at dispatch (here simulated around run_pool) – not from the parent context.

test_worker_crashed_carries_step_name()

The synthesised WorkerCrashedError records the ambient step.

Regression for the “no matching logs found” crash-report gap: the step name must reach the queryable step_name attribute (which the persistence layer writes to error.step_name and crash-report log-collection resolves the run/step logs from), not only the free-text message. The parent stamps it from its ambient context – in the pipeline it is inside error_context(step_name=...) when it launches the pool.

test_worker_crashed_dedups_shared_related_file()

A batch-constant file appears once across many in-flight items.

Two workers in flight against the same reference -> the reference is promoted once, not once per crashed input.

test_worker_crashed_names_inflight_input()

A silent death records the in-flight item(s), not a head sample.

The wrapped worker writes its item into the shared in-flight map before running; a hard os._exit leaves it there, so _worker_crashed can name the culprit in details["crashed_inputs"] (bounded by the worker count).

test_worker_crashed_promotes_related_files()

A silent death promotes the in-flight item to a related file.

So a WorkerCrashedError links straight to the offending file (rendered / FK-resolved), not only a details string.

test_worker_crashed_records_exit_signal()

A native crash records the decoded killing signal.

The tell for OOM/jetsam (SIGKILL) vs. a native crash (SIGSEGV); here a segfault is expected to surface as SIGSEGV. (On Windows the code is an NTSTATUS, not a signal – that path is covered by TestExitSignalDecode.)

test_worker_crashed_records_resources()

A crash records a memory snapshot + the worker count.

N workers vs. total RAM is what makes an OOM/jetsam death easy to judge (paired with a SIGKILL and no native dump from items 4-5).

test_worker_error_carries_config()

A worker error carries its process config (via from_config).

The resolved config lives only at runtime, so recording it is the only way a report shows the settings the step actually ran with.

test_worker_error_carries_reference_files()

A multi-file classifier attaches item and batch constants.

Mirrors the detrending call site: the classifier returns the light curve plus the single photometric reference, so an error carries both.

test_worker_error_carries_related_file()

A worker error carries the item it was processing as a file.

The related_files classifier scopes the item as the ambient related file inside the worker, so an error raised for it (here a FindStarsError) crosses back carrying that file – the datum a config-vs-file-content failure most needs.

test_worker_hard_exit_raises_worker_crashed()

A worker that hard-exits surfaces as WorkerCrashedError, no hang.

ProcessPoolExecutor reports the death as BrokenProcessPool (a bare multiprocessing.Pool would hang here), which run_pool converts into a WorkerCrashedError describing the in-flight inputs instead of blocking the pipeline forever.

class autowisp.tests.test_error_context.TestProcessQueuePropagation(methodName='runTest')

Bases: _ContextTestCase

Process + Queue workers return stamped errors over a queue.

test_all_workers_dead_raises_worker_crashed()

A worker that hard-exits is detected by the parent, not a hang.

manage_astrometry polls is_alive() while waiting on the result queue; when every worker has died without queuing a result it raises a WorkerCrashedError recording the OS exit codes (normalised to the same exit_signal shape as Scheme A), instead of blocking forever.

test_queue_worker_error_round_trips()

A failed queue worker yields a stamped error the parent re-raises.

Mirrors solve_astrometry: a Process worker bootstraps its context, fails with a bare ValueError, and puts capture_for_queue(...) on the queue. The object pulled off is the step’s mapped AutoWISPError (here SolveAstrometryError) carrying the worker subprocess_id, the run snapshot, and the traceback in details; reraise_from_worker then raises it.

class autowisp.tests.test_error_context.TestResourceSnapshot(methodName='runTest')

Bases: TestCase

collect_resource_snapshot is best-effort and never raises.

test_empty_when_psutil_unavailable()

A missing psutil degrades to an empty dict, never a raise.

test_memory_fields_present_and_sane()

psutil is a hard dependency, so the memory fields are present.

class autowisp.tests.test_error_context.TestWorkerEntry(methodName='runTest')

Bases: _ContextTestCase

worker_entry stamps subprocess_id and stays picklable.

test_autowisp_error_passes_through_stamped()

An AutoWISPError keeps its type and gains subprocess_id.

test_bare_exception_wrapped_to_step_subclass()

A bare worker exception becomes the mapped, picklable StepError.

test_inflight_map_cleared_on_error()

A failing task still clears its in-flight entry.

test_inflight_map_tracks_then_clears_item()

The item is in the in-flight map while running, gone after.

(A plain dict stands in for the Manager().dict() proxy; the write/clear logic is the same.)

test_resolve_related_files_callable_returning_many()

A classifier may return several files (item + batch constants).

test_resolve_related_files_from_kind_and_callable()

A FileKind wraps a path item; a callable is used as given.

test_resolve_related_files_is_total()

No classifier, no item, or a bad classifier -> empty, no raise.

test_subprocess_id_not_overwritten_by_later_stamp()

An already-stamped subprocess_id survives a second boundary.

Models an exception stamped by one except that then passes through another on its way out of the same worker: the second stamp is a no-op (stamp_subprocess is idempotent), so the first PID wins rather than being overwritten.

test_success_returns_value()

worker_entry is transparent on the happy path.

class autowisp.tests.test_error_context._ContextTestCase(methodName='runTest')

Bases: TestCase

Base resetting the ambient context to empty before each test.

setUp()

Hook method for setting up the test fixture before exercising it.

autowisp.tests.test_error_context._dr_file(path='/tmp/x.h5', role='input')

A small RelatedFile for related-files assertions.

autowisp.tests.test_error_context._hard_exit_worker(_)

Module-level Pool worker that hard-exits without an exception.

autowisp.tests.test_error_context._instant_exit()

Module-level Process target that hard-exits without an exception.

autowisp.tests.test_error_context._lc_with_reference(item)

A related_files classifier: the item plus a batch-constant ref.

Module-level (so it is picklable to workers) and returns multiple files – the per-item lightcurve and the single photometric reference the whole batch shares – exactly the shape the detrending call site builds via functools.partial.

autowisp.tests.test_error_context._nested_run_pool_worker(item)

Worker that (illegally) tries to launch its own pool.

Used to prove the no-nested-workers guard end-to-end: this runs inside a real worker (so setup_process_map has set in_worker), and the inner run_pool must refuse before spawning anything.

autowisp.tests.test_error_context._noop(*args, **kwargs)

Do-nothing stand-in for the per-image mark_start / mark_end hooks.

autowisp.tests.test_error_context._pool_config(project_home, *, run_id, step='find_stars')

Minimal per-process config for run_pool’s bootstrap.

setup_process_map runs in each worker; pointing it at a temporary project_home keeps its side effects (a throwaway SQLite DB and the per-worker log / stdout-stderr files) inside the temp dir.

autowisp.tests.test_error_context._process_queue_worker(result_queue, config)

Process worker that bootstraps, fails, and queues the stamped error.

Mirrors solve_astrometry.astrometry_process: bootstrap the ambient context from config, catch the failure, and put the result of capture_for_queue (a stamped, picklable error) on the queue rather than raising out of the process.

autowisp.tests.test_error_context._raise_find_stars_error(item)

Module-level Pool worker that raises a StepError.

Must be top-level (not a closure/lambda) so Pool.map can pickle it to the worker.

autowisp.tests.test_error_context._raise_value_error(item)

Module-level Pool worker that raises a bare (non-AutoWISP) error.

autowisp.tests.test_error_context._segfault_worker(_)

Module-level Pool worker that crashes with a real SIGSEGV.

Used to prove faulthandler (armed in setup_process_map) dumps a native traceback into the worker’s own redirected log before it dies.