Repository navigation
feat: trigger.type continuous (0.7.6 preview) is never scheduled; --strict fails on a retargeted overlay patch - #734
Merged
Conversation
…ld, or the rule's description
A `not: {required: [field]}` rule quoted the whole object and the subschema it
must not match: "{'type': 'continuous', 'cron': '0 2 * * *'} should not be valid
under {'required': ['cron']}". It now reads "<field> is not allowed here", as it
already did under sink.catalogAuth, or the rule's own `description` when it
carries one, so a schema can say why and what to do instead.
…eduled A long-running Flink job, or a Kafka Connect or Debezium connector, had no trigger type: every one of them starts runs. `continuous` says the platform keeps the job running, so no scheduler starts it: scheduler synthesis (`fluid generate schedule`, the template-mode schedule) and the acquisition schedule-sync leave the build out, where a connector used to get a nightly 0 2 * * * task, and a contract with no other build has nothing to schedule. The `fluid apply` DAGs already render only schedule triggers. The schema refuses cron, schedule, datasets, event and timetable next to it, and each refusal names the field and the trigger type that takes it. 0.7.5 is unchanged. Borrowed: the name and the "no other trigger" rule from Databricks Jobs' continuous trigger (docs.databricks.com/jobs/triggers.html). Diverged on purpose from Airflow's @continuous (ContinuousTimetable), which starts batch runs back to back: for an always-on job that would loop a no-op DAG.
…her entry The loader logged overlay_patch_retargeted, and a command's --strict reads its own findings, never the log, so `fluid validate --env prod --strict` passed a 0.7.5 overlay that patched the wrong build. `collect_overlay_warnings()` hands the warning to `fluid validate` and `fluid contract-validation`, which report it as their own warning: shown without --strict, failing with it, and listed in `--format json`. The loader logs a warning a block takes at DEBUG only, so the terminal shows it once, and does not mark it as said, so a later load outside a block still logs it. overlay_entry_added stays a log line: an entry only prod adds is often intended, and --strict could not be silenced on it. Borrowed: the context-variable collector already used by build_runners/kafka_connect/iceberg_sink_validation.py (_AT_RUN), and the shape of warnings.catch_warnings(record=True) under CPython 3.14's context_aware_warnings, which keeps one context's warnings out of another's.
…lay patches, readable not-required errors
…h shows it `fluid test --strict` reads ContractValidator's report, so it failed on the overlay warning without printing it: its table had no row for the category, and the loader now logs a collected warning at DEBUG only. It shows as an "Environment overlay" row. Workspace mode (`fluid validate --env prod --strict` with no contract) loaded each product outside the collector and passed; it now lists the warning under the product and fails it under --strict. `fluid validate --strict` says "Validation failed: warnings treated as errors (--strict)" when warnings alone fail it, after a summary that reads valid. The --help example fits 80 columns.
…hrough one predicate The acquisition schedule-sync used is_continuous_build while synthesis and `fluid generate schedule` use is_unscheduled_build; widening the rule would have missed it. The fluid-apply DAG test now gives the continuous trigger a cron, with a schedule-trigger control, so it can fail.
… is its error, --strict overlay coverage
The rule's `if` had no `type: object`, and `required` and `properties` pass on anything else, so a blank `trigger:` (null), a string or a list drew five refusals of a continuous trigger nobody declared beside the real error. The sibling catalogAuth rules already say `type: object`.
…bles escape its ids Under --quiet, which prints errors only, the warning the loader always printed had vanished, and --quiet --strict failed with a count alone. --quiet prints it as a WARNING: line, and under --strict as "ERROR (--strict): <warning>". The warning quotes the contract's ids, which `fluid test`'s "Environment overlay" row and `fluid contract-validation`'s issue table passed to Rich as markup: an id of "[/]" raised MarkupError. Both escape it, the row through test.py's own _detail().
…de schema errors too The ERROR (--strict) lines printed only when warnings alone failed the run, so a contract that also had a schema error showed less under --strict than under --quiet alone, and the second failure came as a surprise after the fix.
…aths that do not know it
github-actions
Bot
added
cli
📄 Documentation ReminderThis PR appears to be missing a documentation reference. Our docs live in a separate repo. Please update the PR description with one of:
See the Contributing Guide for details. |
github-actions
Bot
removed
the
needs-docs
1 of 3 tasks
fas89
added a commit
that referenced
this pull request
Oct 21, 2026
…test.py leaks tests/cli/conftest.py sets FLUID_LOG_LEVEL=DEBUG, and some xdist workers keep it. The CLI subprocesses here inherited it, so the loader's DEBUG copy of a warning the command reports was printed too. The tests that count the warning once then failed with `assert 2 == 1`, as #734's TestValidateStrict did on main's CI run for 1320619. The subprocesses now drop FLUID_LOG_LEVEL, FLUID_DEBUG and FLUID_SHOW_REGISTRATION, as tests/output_ports already do, with API keys as before.
fas89
added a commit
that referenced
this pull request
Oct 22, 2026
… text literally (#738) * fix(console): error lines print exception and contract text literally console_error wrapped its message in [red]...[/red], so Rich parsed the message as markup: a lowercase [word] in it vanished, and a [/] raised MarkupError. The CLI's last-resort handler reports through console_error, so that raise escaped main() as a traceback. main()'s generic handler had the same flaw one step earlier: "[red]Unexpected error: {e}[/red]" raised on the same text and handed it to the fallback. - New _console.literal_text(): a rich Text, never markup, with control characters spelled \xNN. ESC opens an escape sequence and the C1 range holds the one-byte CSI and OSC; printed raw, a message could clear the screen or plant a link. Rich's highlighter broke such sequences up on a top-level str by accident; a Text is not highlighted. - _console.error prints that way, still redacted. None of its 130 call sites passes intended markup (AST scan). - validate.py drops the rich.markup.escape it applied to six messages to survive the old wrapper; kept, it would print a stray backslash. - main(): the Provider error and Unexpected error lines use literal_text, and the --debug traceback prints with markup=False. Text rather than escape: escape doubles a trailing backslash (C:\work\ prints as C:\work\\) when no tag follows it. * fix(cli): fluid test and contract-validation print contract text literally Both reports quote the contract back, and Rich parsed it as markup. A JSON Schema pattern lost its character classes (^[a-z0-9][a-z0-9-]*[a-z0-9]$ printed as ^*$), exposes[orders] lost its id, and an id or expose id holding [/] raised MarkupError, so both commands died with a traceback. Only the overlay row of fluid test and the issue table of contract-validation were escaped (#734). - fluid test: the header panel and every table row go through literal_text; the _detail() escape helper is gone. - contract-validation: the summary panel, each issue's message and path, and a severity outside the styled three go through literal_text. - The Rich-less plain reports, and the soda engine's failed check names and stderr, print with markup=False (the plain fallback strips [word]). JSON, JUnit and the plain report file are unchanged. * fix(console): redact before spelling controls; print soda and --debug text as written Follow-ups from the security review and defect hunt of the two commits before this one. - error() redacts the raw text before it spells control characters, and again after. A spelled control ends in a hex digit, a word character, so "\x00PASSWORD=hunter2" no longer matched the redaction pattern and the secret printed. - New _console.cprint_literal(): cprint with no markup and no :emoji: codes, control characters spelled, still redacted. The soda engine's check names and stderr and the Rich-less plain reports use it; Rich had printed "fe80:🅰️ b:c" with a🅱️ emoji. - main(): the Provider error and Unexpected error lines and the --debug traceback are redacted. The MarkupError used to keep a secret beside a "[/]" off the screen; now that the text prints, it must be redacted. The traceback goes through literal_text as well. - _show_controls handles CR itself, with or without Rich: a CRLF reads as a line feed and a lone CR is spelled. - The CHANGELOG entry now matches the change: contract-validation's issue table was already literal, and only the shared error helper's lines are covered among error output. * fix(console): cprint_literal ignores markup=; pin redaction, CRLF and plain-report spelling Round-2 review follow-ups to the commit before: each of these protections could be removed without a test failing. - main()'s Provider error line is redacted (test with a ProviderError). - error() keeps a CRLF as one line break, with and without Rich. - Both Rich-less plain reports spell control characters in the contract's id and in issue messages. - cprint_literal always turns markup and emoji off, so a caller passing markup= no longer raises TypeError for a duplicate keyword. - The plain-report test in test_contract_validation_ext2 patches cprint_literal, which that report now calls. - The CHANGELOG names the Provider error line and no longer says the reports are redacted. * chore(secrets): refresh .secrets.baseline line numbers Two entries were already stale on main: the Secret Keyword findings in tests/test_cli_test_branches.py (413, now 424) and tests/test_contract_validation_ext2.py (574, now 586) moved when those files grew. CI's detect-secrets step scans only changed files, so it asked for the refresh once this branch touched them.
This was referenced Oct 22, 2026
Merged
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
What and why
Two things #731 and #732 left open.
trigger.type: continuous(0.7.6 preview). A long-running Flink job, or a Kafka Connect or Debezium connector, had no trigger type: schedule, event and the rest all start runs. #731 answered the requester's question with "omittrigger", which still let scheduler synthesis give such a build a nightly0 2 * * *task. Now:continuousjoins$defs.trigger.type. A continuous trigger takes no other trigger:cron,schedule,datasets,eventandtimetableare refused beside it (the rule'sifistype: object, so a blanktrigger:gets one error, not six).util/contract.is_continuous_build, andis_unscheduled_build(= runs outside forge-cli, or continuous). Scheduler synthesis (fluid generate schedule, the template-mode schedule),fluid generate schedule's "nothing to schedule" short-circuit, and the acquisition schedule-sync all use it. Thefluid applyDAGs already render only schedule triggers.format_schema_errorrenders anynot: {required: [field]}error as the rule'sdescription, or<field> is not allowed here, instead of quoting the whole object:cron is not allowed with type continuous: a continuous job runs all the time, so its trigger takes no other trigger. Remove cron, or set type: schedule.AGENTS.md records the convention.--strictfails on a retargeted overlay patch. The loader loggedoverlay_patch_retargeted, and no--strictreads the log.loader.collect_overlay_warnings()(a context-variable collector) hands it to the command, which reports it as its own warning and logs it at DEBUG only, so it shows once:fluid validate --env(one contract, and workspace mode),fluid contract-validation --env,fluid test --env(native engine; an "Environment overlay" row): listed as a warning, and--strictfails on it.--quietstill shows it (WARNING:), and--quiet --strictnames it (ERROR (--strict): …), also beside schema errors.fluid validate --strictsaysValidation failed: warnings treated as errors (--strict)when warnings alone fail it after a summary that reads valid.overlay_entry_addedstays a log line: an entry only prod adds is often intended, and--strictcould not be silenced on it.[/]raisedMarkupError).Borrowed, not built
@continuous(ContinuousTimetable, which starts batch runs back to back): for an always-on job it would loop a DAG that runs nothing.build_runners/kafka_connect/iceberg_sink_validation.py,_AT_RUN), in the shape ofwarnings.catch_warnings(record=True)under CPython 3.14'scontext_aware_warnings.Validation
tests/test_continuous_trigger.py(26) and 16 more intests/test_overlay_keyed_merge.py. Each was red on the commit before its fix (25 onmain, 3 after round 1, 6 after round 2, 1 after round 3).fluid_applytype check, now has a test that kills it.scripts/mypy_strict.py, import-linter (6 kept), license headers, perf budgets (38 passed), bandit. 236 affected test files: 6811 passed. The 20 failures + 2 errors also fail on cleanmainin this venv (environmental; reproduced id-for-id).fluid validateon the requester's Flink contract withtype: continuouspasses; withcronbeside it, one error naming the fix; on 0.7.5, the enum error.fluid generate scheduleon a Flink job and on a continuous connector: "nothing to schedule", no files; the same connector withtype: schedulestill gets its DAG. A 0.7.5 positional overlay:fluid validate --env prodexit 0 with the warning once;--strictexit 1 with the reason;--quiet/--quiet --strict/ workspace mode /fluid test/contract-validationas above.ripwire --quality-delta: the gating rows left are pytest functions as "dead code", per-module test helpers as duplication (house style), the schema's JSON verbosity, and one test class over 60 lines.Reviews
fluid test --strictfailing silently, workspace mode skipping the check, the help example wrapping at 80 columns,docs/schedule.md, one vacuous test.trigger:drawing five continuous refusals,--quiethiding the warning, the markup crash,--quiet --strictbeside schema errors. Final re-verify: all fixed.Documentation
--stricton retargeted overlay patches; merge after this PR)Not in this PR
fluid bundle --env/fluid validate <bundle> --strict(generated pipelines' stage 2) cannot see the warning: the merge is frozen in the bundle. Follow-up: record the collected events in the MANIFESTsourceblock.cron); documented.fluid scaffold-composerwrites a nightly DAG for any contract; the dbt and python runners refusecontinuousas they refuseevent; a DAG deployed before a product went all-continuous is not retired;fluid test's_detail()escape is unused by every other row;--engine sodaignores--strict.🤖 Generated with Claude Code