Upgrade @elastic/charts to 71.8.1 - #2
Closed
angle943 wants to merge 88 commits into
Closed
Conversation
…osdCustom rule (opensearch-project#12215) In joi 17, calling .osdCustom() multiple times on the same schema overwrites the previous rule instead of stacking. This caused: - schema.string({ minLength: 1, maxLength: N }) to only enforce maxLength - schema.string({ minLength, maxLength, validate }) to only enforce validate Combine all string length validations and user validate function into a single osdCustom call. Also prevent the base Type class from adding a separate osdCustom for the validate option when StringType already handles it. Signed-off-by: Justin Kim <jungkm@amazon.com>
…search-project#12214) Signed-off-by: Tomasz Kania <tomasz.kania@pl.ibm.com>
…pensearch-project#12217) Signed-off-by: Qxisylolo <qianxisy@amazon.com>
…search-project#12199) Signed-off-by: Yulong Ruan <ruanyl@amazon.com>
…arch-project#11395) * Fix duplicate shortcut registration for focus_query_bar Renamed the shortcut ID to be unique per instance in the data plugin and distinctive in the explore plugin to prevent crashes when multiple query bars are rendered. Signed-off-by: Mumukshu D.C <phoenixrising656@gmail.com> * Address maintainer feedback: registerKeyboardShortcut prop and useCallback Signed-off-by: Mumukshu D.C <phoenixrising656@gmail.com> * Fix lint errors: extract useCallback, fix prettier formatting - Move useCallback out of conditional useKeyboardShortcut call - Add editorRef to useCallback dependency array - Fix prettier formatting for documentation object indentation - Remove extra space in empty arrow function body Signed-off-by: Abby Hu <abigailhu2000@gmail.com> Signed-off-by: Abby Hu <qingyah@amazon.com> --------- Signed-off-by: Mumukshu D.C <phoenixrising656@gmail.com> Signed-off-by: Abby Hu <abigailhu2000@gmail.com> Signed-off-by: Abby Hu <qingyah@amazon.com> Co-authored-by: Qingyang(Abby) Hu <abigailhu2000@gmail.com> Co-authored-by: Abby Hu <qingyah@amazon.com>
…-project#12223) After grunt was removed, packages/osd-ui-framework/package.json still declares opensearchDashboards.build.intermediateBuildDirectory: 'target', but its 'build' script ('grunt prodBuild') that used to populate target/ was deleted (scripts is now empty). In the production build (@osd/pm buildProductionProjects) the per-project flow is: deleteTarget() wipes target/, buildProject() is now a no-op (no build script, no build targets), copyToBuild() then copies from the empty target/. As a result only package.json is shipped and the committed dist/kui_*.css files are dropped from node_modules/@osd/ui-framework. At runtime core_app.ts serves /node_modules/@osd/ui-framework/dist/{path*} from disk, so the missing kui_*.css returns a JSON 404 and the browser blocks it on an X-Content-Type-Options: nosniff MIME-type mismatch, briefly flashing the fatal-error (red) banner on initial load. The KUI stylesheets are now pre-compiled and committed in dist/, so no build step is required. Remove the stale intermediateBuildDirectory so the package is shipped from its source root (which contains dist/). The build's CleanExtraBuildFiles task strips source .scss/tests/docs from node_modules while preserving .css. Verified with a full distributable build: node_modules/@osd/ui-framework/dist/ now contains all six kui_*.css files (kui_v9_light.css included). Signed-off-by: Justin Kim <jungkm@amazon.com>
…datasets (opensearch-project#12204) Signed-off-by: Yulong Ruan <ruanyl@amazon.com>
…earch-project#12222) Signed-off-by: Qxisylolo <qianxisy@amazon.com>
…helper (opensearch-project#12208) * Replace @amoo-miki/numeral fork with stock @elastic/numeral + BigInt helper The @elastic/numeral dependency was aliased to a personal-scope fork (npm:@amoo-miki/numeral@2.6.0) whose only delta over stock @elastic/numeral 2.5.1 is a BigInt (long-numeral) format branch. This removes the fork in favor of the unmodified upstream package plus a small, well-tested in-repo helper that reproduces the BigInt path byte-for-byte. - package.json + packages/osd-ui-shared-deps/package.json: alias both back to stock @elastic/numeral@2.5.1 (the latter feeds the browser shared-deps bundle) - field_formats/utils/format_bigint.ts: faithful port of numeral's BigInt format path (number/currency/abbreviation/sign/grouping; percent/bytes/ordinal throw on BigInt exactly as the fork did) - converters/numeral.ts: route the isBigInt branch through formatBigInt - format_bigint.test.ts + __fixtures__/numeral_bigint_golden.json: 576-case golden corpus captured from the fork proves bug-for-bug parity, incl. throws Number-path output is unchanged: stock 2.5.1 is byte-identical to the fork for all Number inputs (the fork is just 2.5.1 + the BigInt branch). Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com> Signed-off-by: Joey Liu <jiyili@amazon.com> * Trim comments and drop fork references in BigInt format helper Addresses review feedback: shorten the verbose comments in format_bigint.ts and stop referencing the removed personal-scope numeral fork so future readers aren't confused by a dependency that no longer exists. Describe the helper on its own terms (numeral's BigInt format path) instead. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com> Signed-off-by: Joey Liu <jiyili@amazon.com> --------- Signed-off-by: Joey Liu <jiyili@amazon.com> Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com>
…Engine data sources (opensearch-project#12216) * fix(sample-data): exclude Observability (otel) sample set on AnalyticEngine The sample data list route already limits the available sample sets for an AnalyticEngine (Mustang) data source. It kept both 'logs' and 'otel', but the Observability sample set (otel) cannot be installed on a Mustang domain: its trace index mappings declare 'nested' fields (events/links), which the pluggable data format rejects at index creation: mapper_parsing_exception: nested type is not supported with pluggable data format on field [links] so the install fails with an internal server error ('Unable to install sample data set: Sample Observability Logs, Traces, and Metrics'). Restrict the AnalyticEngine sample set list to 'logs' (Sample web logs), which has no nested fields and installs cleanly. Updates the route's unit test to assert only 'logs' is returned for an AnalyticEngine data source; non-AnalyticEngine sources are unchanged. Signed-off-by: Shenoy Pratik <pshenoy36@gmail.com> * chore: add changelog fragment for opensearch-project#12216 Signed-off-by: Shenoy Pratik <pshenoy36@gmail.com> * Update src/plugins/home/server/services/sample_data/routes/list.ts Co-authored-by: Joshua Li <joshuali925@gmail.com> Signed-off-by: Shenoy Pratik <sgguruda@amazon.com> --------- Signed-off-by: Shenoy Pratik <pshenoy36@gmail.com> Signed-off-by: Shenoy Pratik <sgguruda@amazon.com> Co-authored-by: Joshua Li <joshuali925@gmail.com>
opensearch-project#12258) * fix(data_source): prevent privilege escalation via unauthorized data source usage The getClient(dataSourceId) context provider passed the calling user's scoped SavedObjects client directly to configureClient/configureLegacyClient, which then fetched and decrypted the data source's stored credentials with no authorization check on whether the caller was permitted to USE that data source. Any authenticated user with read access to a tenant could use any data source's stored credentials to execute requests as that data source's identity, bypassing index-level, document-level, and field-level security (CVE pending). Fix (two layers): Layer 1 — Strip credentials from read APIs in DataSourceSavedObjectsClientWrapper. The wrapper previously only intercepted write operations (create/update). get(), find(), and bulkGet() now strip auth.credentials from data-source saved objects before returning them to any caller, so encrypted credential material is never exposed through the SavedObjects API. Layer 2 — Internal repository for credential access in getClient. DataSourcePlugin.start() creates an internal SavedObjects repository that bypasses the credential-stripping wrapper. configureClient and configureLegacyClient now perform a two-step fetch: 1. getDataSource(id, scopedClient) — access check; throws 404/Forbidden if the calling user's tenant/workspace does not include this data source. 2. getDataSourceInternal(id, internalRepo) — fetches with full encrypted credentials only after step 1 passes. internalSavedObjects is optional in DataSourceClientParams for backward compatibility; callers that do not provide it (e.g. test-connection routes, existing unit tests) retain the previous single-fetch behavior with no change in call count or semantics. Affects: open-source self-managed deployments with data_source.enabled: true. Not affected: managed service (token_exchange auth model, no stored credentials). Signed-off-by: Zilong Xia <zilongx@amazon.com> * fix: address lint errors (prettier formatting, array type style) Signed-off-by: Zilong Xia <zilongx@amazon.com> --------- Signed-off-by: Zilong Xia <zilongx@amazon.com>
…12234) Signed-off-by: Tomasz Kania <tomasz.kania@pl.ibm.com>
* fix auto creation field detection Signed-off-by: Adam Tackett <tackadam@amazon.com> * address comment Signed-off-by: Adam Tackett <tackadam@amazon.com> --------- Signed-off-by: Adam Tackett <tackadam@amazon.com> Co-authored-by: Adam Tackett <tackadam@amazon.com>
* fix side nav Signed-off-by: Adam Tackett <tackadam@amazon.com> * address comment Signed-off-by: Adam Tackett <tackadam@amazon.com> --------- Signed-off-by: Adam Tackett <tackadam@amazon.com> Co-authored-by: Adam Tackett <tackadam@amazon.com>
…pensearch-project#12219) Signed-off-by: Divya Madala <divyaasm@amazon.com>
…12266) * chore(deps): hono 4.12.25 Signed-off-by: Tomasz Kania <tomasz.kania@pl.ibm.com> * chore(deps): markdown-it 14.2.0 Signed-off-by: Tomasz Kania <tomasz.kania@pl.ibm.com> * chore(deps): js-yaml 4.2.0 Signed-off-by: Tomasz Kania <tomasz.kania@pl.ibm.com> * chore(deps): ws 8.21.0, 7.5.11 Signed-off-by: Tomasz Kania <tomasz.kania@pl.ibm.com> * chore(deps): form-data 4.0.6 Signed-off-by: Tomasz Kania <tomasz.kania@pl.ibm.com> * chore(deps): tar 7.5.16 Signed-off-by: Tomasz Kania <tomasz.kania@pl.ibm.com> --------- Signed-off-by: Tomasz Kania <tomasz.kania@pl.ibm.com>
…alcite proxy routes (opensearch-project#12255) Signed-off-by: Hanyu Wei <weihanyu@amazon.com> Co-authored-by: Hanyu Wei <weihanyu@amazon.com>
…rds) (opensearch-project#12272) - Replace old backport workflow (VachaShah/backport + GitHub App) with reusable workflow - Remove delete_backport_branch.yml (now handled by reusable workflow) Signed-off-by: Peter Zhu <zhujiaxi@amazon.com>
…ari (opensearch-project#12231) Signed-off-by: Yulong Ruan <ruanyl@amazon.com>
…pensearch-project#12221) visualization snapshot Signed-off-by: Yulong Ruan <ruanyl@amazon.com>
… column identifier (opensearch-project#12218) * clean up uniqueValuesCount/validValuesCount Signed-off-by: Yulong Ruan <ruanyl@amazon.com> * fix lint Signed-off-by: Yulong Ruan <ruanyl@amazon.com> --------- Signed-off-by: Yulong Ruan <ruanyl@amazon.com>
Signed-off-by: Tomasz Kania <tomasz.kania@pl.ibm.com>
…ersion (opensearch-project#12270) * chore: use major version in .nvmrc to avoid requiring exact Node.js version .node-version pins the exact runtime version used by the build system for downloading Node.js binaries; .nvmrc only needs to guide developers to the correct major version via nvm. - Change .nvmrc from exact version to major version (22) - Update build config to read .node-version for exact version Signed-off-by: Yulong Ruan <ruanyl@amazon.com> * Update DEVELOPER_GUIDE.md for nodejs version usage Signed-off-by: Yulong Ruan <ruanyl@amazon.com> --------- Signed-off-by: Yulong Ruan <ruanyl@amazon.com>
…12233) Signed-off-by: Lin Wang <wonglam@amazon.com>
…earch-project#12280) The Data Table visualization was missing from the "Create New Visualization" type picker. createBaseVisualization() was being called after `await core.getStartServices()`, which only resolves once the plugin's start phase runs. Since vis types must be registered during the setup phase (before the registry is read to build the picker), the type was registered too late and silently never appeared. Move createBaseVisualization() ahead of the getStartServices() await so registration happens synchronously during setup. The expressions renderer registration still awaits start services since it genuinely needs the CoreStart contract. Signed-off-by: Yulong Ruan <ruanyl@amazon.com>
…ss CVEs (opensearch-project#12282) * chore(deps): postcss-selector-parser from 6.0.10 to 6.1.4 Signed-off-by: Tomasz Kania <tomasz.kania@pl.ibm.com> * chore(deps): remove not need it resolutions Signed-off-by: Tomasz Kania <tomasz.kania@pl.ibm.com> * chore(deps): dompurify 3.4.11 Signed-off-by: Tomasz Kania <tomasz.kania@pl.ibm.com> * chore(deps): pixelmatch from 5.2.1 to 5.3.0 Signed-off-by: Tomasz Kania <tomasz.kania@pl.ibm.com> * chore(deps): pngjs 7.0.0 Signed-off-by: Tomasz Kania <tomasz.kania@pl.ibm.com> --------- Signed-off-by: Tomasz Kania <tomasz.kania@pl.ibm.com>
…pensearch-project#12281) When a slash command is registered or unregistered at runtime, the command suggestion dropdown now updates immediately. Previously, suggestions only refreshed on user input change, causing stale entries to remain visible if a command was dynamically removed. Adds an onChange listener to SlashCommandRegistry that notifies the useCommandMenuKeyboard hook to re-evaluate suggestions. Signed-off-by: Lin Wang <wonglam@amazon.com>
…ver page (opensearch-project#12278) Signed-off-by: tq0905 <1352711780@qq.com>
* chore(build): upgrade Rspack to v2
Disable import/no-unresolved for @rspack/core require
@rspack/core v2 is published as a pure ESM package ("type": "module"
with no CJS entry in exports). ESLint's import resolver cannot resolve
ESM-only packages in a require() context. Node.js 22 handles this at
runtime via native require(esm) support, so the import works correctly
despite the resolver limitation.
Signed-off-by: Yulong Ruan <ruanyl@amazon.com>
* update test snapshots
Signed-off-by: Yulong Ruan <ruanyl@amazon.com>
* update failing snapshots
Signed-off-by: Yulong Ruan <ruanyl@amazon.com>
---------
Signed-off-by: Yulong Ruan <ruanyl@amazon.com>
… plugin registration) (opensearch-project#12362) * fix: don't terminate lint on ESLint 10 concurrency warning ESLint 10 emits an advisory ESLintPoorConcurrencyWarning when `concurrency: 'auto'` (src/dev/run_eslint.js) selects multithreading for a small workload, such as linting a single plugin directory. The process-warning handler in setup_node_env/exit_on_warning.js treats any unlisted Node process-warning as fatal and calls process.exit(1), so `yarn lint` exited 1 even with zero lint errors, breaking CI for downstream plugins. Add ESLintPoorConcurrencyWarning to the ignore list alongside the other advisory warnings so it no longer terminates the process. Real lint errors still exit 1. Signed-off-by: SuZhou-Joe <suzhou@amazon.com> * feat: share plugin registration via @elastic/eslint-config-kibana Plugins that consume the shared flat config should not have to know which ESLint plugin owns a rule or re-register it. In ESLint 10 flat config, `plugins` registration is scoped per matched file and is not inherited between config objects, so a plugin spreading the shared config and referencing e.g. `@typescript-eslint/no-explicit-any` in its own block would fail with 'Could not find plugin' unless it re-required and re-registered the plugin. Register every bundled plugin once in an unscoped config object (`globalPluginRegistration`) placed first in the exported array, so downstream rule references resolve at any file scope with no per-plugin boilerplate. - plugins.js: single source of truth for plugin instances + global registration - eui.js: opt-in EUI config block (EUI is not part of the base config) - extras.js: aggregates { eui, plugins, restrictedGlobals } behind one subpath - eslint.config.js: sources instances from plugins.js; global registration first - package.json: add exports map (., ./eslint.config.js, ./extras); align peers - plugin generator template + README: use the spread pattern, no registration Signed-off-by: SuZhou-Joe <suzhou@amazon.com> * style: apply prettier 3 formatting to workspace/server/utils These files (added in opensearch-project#12343) carry nested-ternary indentation that Prettier 3 reformats but that landed before this branch surfaced it under the ESLint 10 / Prettier 3 lint. Pure formatting change; no logic change. Signed-off-by: SuZhou-Joe <suzhou@amazon.com> --------- Signed-off-by: SuZhou-Joe <suzhou@amazon.com>
…0.2 tar 7.5.19 (opensearch-project#12364) * chore(deps): hono 4.12.28 immutable 5.1.9 js-yaml 4.3.0 linkify-it 5.0.2 tar 7.5.19 Signed-off-by: Tomasz Kania <tomasz.kania@pl.ibm.com> * fix: formatting issues after eslint 10 merge Signed-off-by: Tomasz Kania <tomasz.kania@pl.ibm.com> --------- Signed-off-by: Tomasz Kania <tomasz.kania@pl.ibm.com>
…ensearch-project#12368) Signed-off-by: yubo luo <yubonluo@amazon.com>
…project#12345) --------- Signed-off-by: Yulong Ruan <ruanyl@amazon.com>
…pensearch-project#12372) Signed-off-by: Hailong Cui <ihailong@amazon.com>
opensearch-project#12346) When starting a new conversation while tool calls were in pending/executing state, the input remained disabled because stale toolCallStates from the previous run persisted in the global AssistantActionService singleton. Add eventHandler.clearState() to handleNewChat to drain all pending tool calls and reset the input-blocking hasActiveToolCalls check. Signed-off-by: Lin Wang <wonglam@amazon.com>
… dashboard (opensearch-project#12376) resoved opensearch-project#12337 Signed-off-by: Yulong Ruan <ruanyl@amazon.com>
* Refactoring get_query_with_source * Fix mangled regex and handle quotes * Fix prettier lint test issue * Fix Regex * Fix backtick edge case Signed-off-by: Ajimelec Gonzalez <ajimelec@amazon.com>
…t#12319) * feat(ppl-lint): add 3 structural rules (STRUCT-A) Port invalid-capture-group-name, unsupported-window-function-in-eventstats, and multisearch-min-subsearch from poc-ppl-linter-v3. All three are structural rules with no context dependency — they fire the moment they merge. - invalid-capture-group-name: flags Java-invalid regex group names in rex/parse/grok, with quick-fix suggestions (sanitize or convert Python opener) - unsupported-window-function-in-eventstats: flags rank/dense_rank/ntile/etc in eventstats/streamstats (only row_number is supported) - multisearch-min-subsearch: flags multisearch with fewer than 2 subsearches (runtime-only, minVersion 3.4.0) Also adds DiagnosticFix interface to diagnostic.ts (additive, no consumers yet on this branch — enables the capture-group quick-fix). Signed-off-by: Hanyu Wei <weihanyu@amazon.com> * feat(ppl-lint): add 4 more structural rules + address review feedback Add four rules ported from poc-ppl-linter-v3: - disabled-join-type (warning): flags right/full/cross join types that are disabled unless plugins.calcite.all_join_types.allowed is set. Reachable on the compiled surface through the type=<keyword> option form; suppressed when the allJoinTypesAllowed setting is on. - dedup-consecutive-unsupported (Calcite warning): flags dedup consecutive=true, which is not natively supported on Calcite and relies on engine fallback. - union-min-datasets (runtime-only, Calcite, minVersion 3.7.0): flags union with fewer than two datasets. No-op on the compiled surface (unionCommand absent). - replace-wildcard-asymmetry (runtime-only, Calcite, minVersion 3.4.0): flags a replace whose pattern and replacement have asymmetric, non-zero wildcard counts. No-op on the compiled surface (replacePair absent). Also address PR review feedback on invalid-capture-group-name: - Emit a dedicated message for an empty group name "(?<>...)" instead of the generic "Invalid capture group name """ text. - Build the capture-group opener regex per call rather than sharing a global regex, removing any lastIndex state hazard. - Add regression tests asserting the diagnostic and quick-fix ranges land on the exact column of the group name and the P character (the reported offset concern is not a real defect; the offsets are measured consistently from the token, quotes included). Each rule is wired into detector_registry, rules_catalog.json, and the query_enhancements uiSettings defaults. 134 lint tests pass; tsc and eslint clean. Signed-off-by: Hanyu Wei <weihanyu@amazon.com> * fix(ppl-lint): widen empty-capture-group range + grok regression test Address the persistent-review focus areas on the STRUCT-A rules: - invalid-capture-group-name: for an empty name "(?<>...)", the diagnostic range previously covered only the closing ">"; widen it one column left so it spans the whole empty "<>" pair. Add a regression test asserting the range. - Add a regression test confirming grok semantic names (typed "%{NUMBER:x:int}" and dashed "%{DATA:my-field}") are not flagged — the (?<name>) opener scan never matches %{...} syntax, so there is no false positive. - replace-wildcard-asymmetry: document why countUnescapedWildcards runs on the raw getText() literal (it lints the escapes as written in the editor). 135 lint tests pass; tsc and eslint clean. Signed-off-by: Hanyu Wei <weihanyu@amazon.com> * refactor(ppl-lint): drop invalid-capture-group-name rule (quick-fix lands with opensearch-project#12298) The invalid-capture-group-name rule was the only rule here that emitted a DiagnosticFix, but this PR ships no code-action provider to render it — that machinery (fix_registry + code_action_provider + registerCodeActionProvider) lives on opensearch-project#12298. Rather than carry a rule whose quick-fix cannot surface and duplicate the DiagnosticFix interface across two PRs, drop the rule and the unused fix data model here; the rule can re-land alongside its quick-fix on opensearch-project#12298. Removes the detector, its tests, the catalog entry, the uiSettings default, and the DiagnosticFix interface + Diagnostic.fix field from diagnostic.ts. This PR now adds 6 structural rules. 123 lint tests pass; tsc and eslint clean. Signed-off-by: Hanyu Wei <weihanyu@amazon.com> * fix(ppl-lint): wire local-cluster version fallback and calcite settings into lint context buildPPLLintContext now falls back to the grammar cache's resolved version when the dataset metadata does not carry one (common on local clusters), fixing Calcite-gated rules staying disabled on local datasets. A new calcite_settings_cache module fetches and caches per-datasource Calcite cluster settings, wiring allJoinTypesAllowed into the lint context so that the disabled-join-type rule can be suppressed when the cluster has all join types enabled. Both caches notify the editor on resolution so lint context is refreshed without a user-triggered revalidation. Signed-off-by: Hanyu Wei <hanyuwei@amazon.com> Signed-off-by: Hanyu Wei <weihanyu@amazon.com> * chore(ppl-lint): fix prettier formatting for eslint 10 / prettier 3.9.4 Signed-off-by: Hanyu Wei <weihanyu@amazon.com> * fix(ppl-lint): address Joshua + Eric review on 6 structural rules Fixes the three findings from Joshua's Claude Code review plus a gating bug and a test-coverage gap surfaced while re-verifying Eric's comments. J3 / P0-a — union-min-datasets fired an error on every valid mid-pipeline union (`... | union [subsearch]`), where the upstream pipeline is the implicit first dataset. Only enforce the >=2 check when the unionCommand is query-initial (parent rule `pplCommands`); stay silent mid-pipeline. Uses a positive gate so an unresolved parent defaults to not firing. J2 / P0-b — disabled-join-type used a recursive joinType descendant search, which on nested joins (`join type=inner b [ ... | join type=cross d ]`) reported the nested keyword twice and masked the outer join's own type. Read joinType only from each join's own direct `joinOption` children, matching the grammar (`joinCommand : JOIN (joinOption)* ...`). The sqlLikeJoinType prefix form is unchanged. J1 / P0-c — all six detectors hardcoded `message` and ignored config.message. Static rules now use config.message; the three interpolated rules move their dynamic detail into hoverFacts (joinType, windowFunction, wildcard counts) and the hover card renders those facts. P1 — the Calcite settings cache warmup was gated behind query:enhancements:runtimePplGrammar, so with the flag off the compiled-surface disabled-join-type rule could never be suppressed by allJoinTypesAllowed. The warmup handler now subscribes unconditionally (the grammar cache self-gates on the flag internally); only the runtime validation provider stays gated. P2 — add a product-path plumbing test that drives the runtime-only rules through getBundledCatalog + the detector registry + runLint, so a catalog disable, runtimeOnly-filter break, registry unregistration, or rule-name divergence is caught (the prior tests exercised only bare detectors). Also map the `@osd/monaco/ppl-lint` subpath to its Monaco-free source barrel in the Jest config. The `@osd/monaco` barrel is globally jest.mock()'d, but the subpath resolved to `target/` (blocked by modulePathIgnorePatterns), so runtime_lint.test.ts could never load without a build. The mapped source barrel depends only on antlr4ng/@osd/antlr-grammar/semver. Adds nested-join, mid-pipeline-union, catalog-message, and settings-warmup tests. Signed-off-by: Hanyu Wei <weihanyu@amazon.com> --------- Signed-off-by: Hanyu Wei <weihanyu@amazon.com> Signed-off-by: Hanyu Wei <hanyuwei@amazon.com> Co-authored-by: Hanyu Wei <weihanyu@amazon.com>
…pace, or tab (opensearch-project#12383) * fix(workspace): split configUsers and configGroups string by comma or space When configUsers or configGroups is a string (e.g. from YAML config), split it by commas or whitespace so multiple values are handled correctly. Signed-off-by: Hailong Cui <ihailong@amazon.com> * test(workspace): add tab delimiter test cases for configUsers and configGroups splitting Signed-off-by: Hailong Cui <ihailong@amazon.com> --------- Signed-off-by: Hailong Cui <ihailong@amazon.com>
* feat: add user/workspace setting page Signed-off-by: yubonluo <yubonluo@amazon.com> * update getAllUserProvidedWithScope name and test snapshot Signed-off-by: yubonluo <yubonluo@amazon.com> * optimize the code Signed-off-by: yubonluo <yubonluo@amazon.com> * use saveobject as the permission control source Signed-off-by: yubonluo <yubonluo@amazon.com> --------- Signed-off-by: yubonluo <yubonluo@amazon.com>
…#12380) * fix(chat): serialize tool-result run to avoid ConcurrencyException (opensearch-project#11881) A frontend-tool round-trip dispatches a second run once the tool executes in the browser. When a parallel backend tool is still resolving, the first run still holds the thread, so dispatching the tool-result run hits the agent server's per-thread concurrency guard and throws ConcurrencyException. Wait for other active runs to drain (bounded, best-effort) before dispatching the tool-result run, serializing the two runs instead of racing them. Fixes opensearch-project#11881 Signed-off-by: cicimiao31 <cicimiao31@gmail.com> * Prioritize AG-UI routing over ML Commons in proxy Signed-off-by: YiyuanMiao <miaoyiyuan31@gmail.com> --------- Signed-off-by: cicimiao31 <cicimiao31@gmail.com> Signed-off-by: YiyuanMiao <miaoyiyuan31@gmail.com> Co-authored-by: cicimiao31 <cicimiao31@gmail.com>
…ect#12375) - Dashboard URLs now store only variable selections as { id, current }. - Full variable definitions stay in saved dashboard variablesJSON. - URL selections are hydrated back onto saved/current variables by id. - Deleted or unknown URL variables are ignored. --------- Signed-off-by: Yulong Ruan <ruanyl@amazon.com>
…mponent 0.5.0 (opensearch-project#12366) Signed-off-by: Tomasz Kania <tomasz.kania@pl.ibm.com>
…rch-project#12333) Renders tool results by shape (JSON, {text=...} wrappers, tables, markdown, plain text). Centralizes truncation in CodeBlock only; MarkdownBlock never truncates. .text extraction requires the object have no other keys, to avoid dropping sibling fields. Signed-off-by: Lin Wang <wonglam@amazon.com>
…saved_search_in_dashboards -> ciGroup 16) (opensearch-project#12389) Move two mis-bucketed explore Cypress specs into feature-coherent ciGroups so each group maps to a single feature area: - ai_editor.spec.js: 10 -> 19 (ciGroup 19 groups AI features alongside results_summary.spec.js) - saved_search_in_dashboards.spec.js: 12 -> 16 (ciGroup 16 groups dashboard integration tests: add vis/log to dashboard, download CSV from dashboard) Both are pure git renames. Source and target directories are at the same relative depth, so imports are unchanged. The ciGroup scripts in package.json glob each bucket directory (explore/<n>/*.spec.js), so they pick up the moved files automatically with no workflow or package.json changes required. Histogram specs are intentionally left in ciGroup 14 as Discover features. Signed-off-by: SuZhou-Joe <suzhou@amazon.com>
Signed-off-by: Adam Tackett <tackadam@amazon.com> Co-authored-by: Adam Tackett <tackadam@amazon.com>
…-project#12379) Signed-off-by: Ritvi Bhatt <ribhatt@amazon.com>
* chore(deps): jest 30.4.2 Signed-off-by: Tomasz Kania <tomasz.kania@pl.ibm.com> * chore: jest-30-migration-recipe codemod Signed-off-by: Tomasz Kania <tomasz.kania@pl.ibm.com> * test: adjust tests to jest 30 and jsdom 26 part1 Signed-off-by: Tomasz Kania <tomasz.kania@pl.ibm.com> * test: add jest-location-mock Signed-off-by: Tomasz Kania <tomasz.kania@pl.ibm.com> * test: adjust Signed-off-by: Tomasz Kania <tomasz.kania@pl.ibm.com> * fix: add MIT-0 license as allowed, tests adjustments Signed-off-by: Tomasz Kania <tomasz.kania@pl.ibm.com> * fix: dependencies cleanup Signed-off-by: Tomasz Kania <tomasz.kania@pl.ibm.com> * test: adjust after rebase Signed-off-by: Tomasz Kania <tomasz.kania@pl.ibm.com> --------- Signed-off-by: Tomasz Kania <tomasz.kania@pl.ibm.com>
) * chore(ppl-lint): re-trigger CI (flaky cypress ciGroup23Explore) Signed-off-by: Hanyu Wei <weihanyu@amazon.com> * feat(ppl-lint): command-typo quick-fix + unknown-field rule Adds the quick-fix subsystem with two fix producers, exercising the lightbulb on both marker channels end-to-end: 1. Command-typo suggestion (syntax channel) — replaces ANTLR's noisy "mismatched input ... expecting {N keywords}" with "Did you mean 'where'?" plus a one-click rewrite. Fires on both the compiled-worker path (ppl_error_listener) and the runtime ParserInterpreter path (new PPLCommandErrorListener). 2. field-validation lint rule (lint channel) — flags unknown field references and the Splunk-style `grok field=body` shape error, each with a quick-fix ("Replace with '<field>'" / "Remove field="). Quick-fix infra: fix_registry side tables (lint + syntax channels, keyed by the marker fields Monaco's MarkerService preserves) and a code_action_provider that routes by marker source. markerFixKey moves to fix_registry (canonical owner); hover_registry re-exports it. Field context: minimal fields-only plumbing — LintFieldsCache, 3-arg buildPPLLintContext, and a loadFields effect in both hosts (query_editor + explore use_query_panel_editor) that walks indexPattern.fields into a Set. No server route, no version coupling. field-validation self-suppresses until fields load. The plumbing makes the sibling needsContext rules ~1 catalog line each; omitting them here is a scoping choice. Catalog/ui_settings/doc_links parity kept in lock-step (field-validation added to all three). GeneralErrorListener widened backward-compatibly (trailing optional RecognitionException + optional code/fix) so the SQL /DQL/autocomplete callers are unaffected. Signed-off-by: Hanyu Wei <weihanyu@amazon.com> * ci: retrigger after transient Cypress download timeout Signed-off-by: Hanyu Wei <weihanyu@amazon.com> * fix(ppl-lint): don't flag the source/index keyword as an unknown field Live verification against a sub-3.6 cluster (OpenSearch 2.19) surfaced a false positive: on the compiled-simplified grammar, `source=idx` / `index=idx` parses the leading `source`/`index` keyword into a fieldExpression (the runtime >=3.6 grammar instead parses it as a fromClause, which field-validation already excludes). Because Group D wires `fields` into the compiled-worker lint context, field-validation now runs on that surface and emitted a spurious `Unknown field "source"` on essentially every source-first query against a sub-3.6 cluster — the dominant query shape in the editor's compiled-fallback path. Skip the `source`/`index` fromClause keyword in the existence pass. The shape pass already defers on the compiled surface, and >=3.6 clusters are unaffected (the runtime grammar excludes the fromClause). Real unknown fields on source-first queries are still flagged. Adds 4 regression cases. Signed-off-by: Hanyu Wei <weihanyu@amazon.com> * fix(ppl-lint): validate eval RHS + guard cast target type The field-validation existence pass excluded the entire `evalClause` subtree via ancestor-matching, so a typo in an eval value expression (`eval x = nonexistent + 1`) was never flagged even though the eval target is already protected by `createdFields`. Drop `evalClause` (and the dead `renameClasue`/`renameClause` entries) from the excluded-ancestor set so the RHS is walked like a `where` clause; the created-field name slot stays safe via `createdFields`. Separately, `collectCreatedFields` added the node after any `AS` terminal to the created-field set, so `cast(field AS int)` registered the target type name as a field — silently accepting a later reference to it. Skip the `AS` slot when the following node is a `convertedDataType`. Also convert the touched `stack.pop()!` non-null assertions to the guard form. Signed-off-by: Hanyu Wei <weihanyu@amazon.com> * fix(ppl-lint): scope alternate-source field creation to its own pipeline `buildPipelineShape` collected created fields from every command stage, including stages nested inside an alternate-source subtree (`append [search ...]`, subsearch, lookup, appendcol, union). A field created by an eval inside `append [...]` therefore leaked into the outer pipeline's known-field set, silently accepting a downstream reference to it. Prune alt-source stages when collecting created fields, reusing the same `collectAlternateSourceSubtrees` set the reference pass already prunes against. Extract the ancestor-walk into a shared `isInsideAltSource` helper so `buildPipelineShape` and `head_without_sort` prune identically instead of each inlining the same closure. Signed-off-by: Hanyu Wei <weihanyu@amazon.com> * fix(ppl-lint): don't suggest a command for sub-threshold-length tokens `suggestCommand` used an edit threshold of 1 for short tokens, so a single stray character (`source=logs | a`) was distance 1 from any 2-char command (`ad`, `ml`) and got "corrected". The structured suggestion then replaced ANTLR's real diagnostic and attached a misleading quick-fix while the user was mid-typing a fresh command. Add a length floor: a token must be strictly longer than the edit threshold to be considered a typo — rewriting 100% of a token to reach a command is a guess, not a correction. Signed-off-by: Hanyu Wei <weihanyu@amazon.com> * fix(ppl-lint): honor configured severity in field-slot shape pass The existence pass emitted at `config.severity`, but the field-slot shape pass hardcoded `severity: 'error'`. A user who set field-validation to `warning` still saw red errors from the shape pass. A rule has one catalog entry and one user-facing severity toggle, so the shape pass must follow it; use `config.severity`. Signed-off-by: Hanyu Wei <weihanyu@amazon.com> * fix(ppl-lint): only skip source/index keyword on the compiled surface The existence pass unconditionally skipped any field reference whose text was `source`/`index`, so a real field literally named `source` or `index` was never validated on any surface. The skip is a compiled-simplified-grammar workaround: there `source=idx` misparses the keyword into a fieldExpression. On the runtime bundle `source=idx` is an excluded fromClause, so a fieldExpression named `source`/`index` is a genuine reference. Gate the skip on `grammarSurface !== 'runtime-bundle'`, which keeps the workaround on the compiled and test/fallback paths (no surface set) while restoring real validation of those field names on the modern surface. Signed-off-by: Hanyu Wei <weihanyu@amazon.com> * fix(ppl-lint): guard syntax highlighting against stale async results `processLintHighlighting` already drops a slower, earlier pass whose response resolves after a newer one (generation counter + content/dispose checks). `processSyntaxHighlighting` had no such guard: it awaited the validation result and wrote markers/fixes unconditionally. Multiple triggers (dataset change, loadFields resolve, content change) call revalidatePPLModel, so two in-flight validations can race and the stale one resolving last clobbers newer markers — leaving a stale command-typo fix range pointing into outdated text. Add a dedicated `syntaxGenerations` guard mirroring the lint path, with matching cleanup on model dispose and on provider teardown. Signed-off-by: Hanyu Wei <weihanyu@amazon.com> * fix(ppl-lint): robust join-alias collection across grammar surfaces `collectJoinAliases` used `findAllChildrenByRule` (direct children only) to find the `qualifiedName` under each `sideAlias`. If a grammar surface nests that `qualifiedName` below an intermediate rule, the alias set comes back empty and every join-alias reference (`| where l.response = 200`) is false-flagged. The join-alias tests only run on the compiled surface, so this risk is untested on the runtime surface today. Switch to `findAllDescendantsByRule`. A `sideAlias` wraps exactly one alias identifier, so descending cannot over-collect; the change is strictly more robust across surfaces. Signed-off-by: Hanyu Wei <weihanyu@amazon.com> * refactor(ppl-lint): unify edit-distance on Damerau-Levenshtein The command suggester used Damerau-Levenshtein (transpositions cost 1) while the field suggester used plain Levenshtein (transpositions cost 2, compensated by a wider threshold) — two near-identical sweeps that differed only by which typo class they ranked well. Extract a shared `edit_distance` module exposing `damerauLevenshtein` and a `nearestWithinThreshold` helper, and route both `suggestField` and `suggestCommand` through it. Delete the plain `levenshtein` implementation; both callers now rank transposition typos closer at no extra cost. `command_suggestion` re-exports `damerauLevenshtein` so existing importers are unaffected. Behavior is preserved (verified by the existing suggester tests, including the distance-0 case-preference case). Signed-off-by: Hanyu Wei <weihanyu@amazon.com> * chore(ppl-lint): collision tripwire + hot-path tidy + shared field extraction Three non-behavioral cleanups bundled: - Add a dev-only `console.warn` in the lint marker-capture loop when a markerFixKey would be overwritten with a different fix. Today separate fix tables and suppressContained prevent collisions, so this is a latent tripwire; the dead branch is stripped from production bundles. - Replace the per-keystroke `new Set([...fields, ...createdFields])` merge in the field-existence pass with a two-set membership test (`isKnown`); the cold suggestion path materializes the union inline. - Extract `extractFieldNames` into lint_context_builder and route both the data query_editor and the explore use_query_panel_editor loadFields effects through it, replacing two copies of the same loop. The data copy gains the `!indexPattern` guard the explore copy already had. Signed-off-by: Hanyu Wei <weihanyu@amazon.com> * fix(ppl-lint): lint best-effort on the runtime surface even when parsing errors `buildRuntimeTree` discarded the entire parse tree whenever the runtime `ParserInterpreter` reported any syntax error, so `lintRuntimePPLQuery` returned no diagnostics for the whole query. This silently disabled all linting for queries the deserialized runtime ATN can't fully parse even though they are semantically valid to the engine (e.g. `eval x = <field> + 1`), and for any otherwise-valid query with a trailing unparseable command. ANTLR's error recovery still produces a usable partial tree, and the lint rules already walk it best-effort. The compiled fallback path (`PPLLanguageAnalyzer.lint`) never discards a tree, so the runtime path was the outlier. Keep the recovered tree; only a thrown exception (no tree at all) now suppresses linting. Browser-verified against a live 3.8 runtime grammar bundle: `eval x = nonexistent + 1` now flags the unknown RHS field where it previously produced nothing. Signed-off-by: Hanyu Wei <weihanyu@amazon.com> * fix(ppl-lint): register extraction-command created fields The field-validation existence pass unions index fields with fields created upstream, but collectCreatedFields never registered the fields that grok/parse/rex/patterns/spath produce. Any downstream reference to an extracted field was therefore false-flagged as an unknown field once a real index field set was present. Add a dependency-free pattern_fields module that extracts created names from a pattern string (grok %{SYNTAX:name}; Java (?<name>) groups filtered by the engine's charset rules — no underscore for parse/rex), and three branches in collectCreatedFields: - grok/parse/rex: names from the pattern string literal - patterns: NEW_FIELD name + default patterns_field + companion tokens - spath: OUTPUT name or path-derived field (INPUT left validated) Because buildPipelineShape here scopes collectCreatedFields to non-alternate-source stages, a field extracted inside an `append [search ...]` subquery does NOT leak into the outer known-field set — an outer reference to it is correctly flagged (regression test added). Verified live against the Calcite 2.19 and runtime 3.6 engines. The two disagree on patterns output naming (2.19 honors NEW_FIELD and emits a tokens column; 3.6 ignores NEW_FIELD and always emits patterns_field), so the patterns branch registers the union to stay correct on both. field_validation.ts is unchanged. No grammar changes. Signed-off-by: Hanyu Wei <weihanyu@amazon.com> * fix(ppl-lint): don't register a join table alias as a created field `collectCreatedFields` treated any node after an `AS` terminal (except a cast's `convertedDataType`) as a created field. On a top-level `join departments AS d`, the `AS` sits under `tableSourceClause` and binds `d` as a *table* alias, not a column on the outer source — so `d` leaked into the pipeline's known-field set and silently suppressed a real "Unknown field" warning on a downstream reference to a field named `d`. (The `lookup ... AS d` shape was already safe: lookupCommand is pruned by collectAlternateSourceSubtrees, so collectCreatedFields never runs on it.) Skip an `AS` whose immediate container is a table/source-alias context, reusing the same vocabulary field_validation's existence pass already excludes (tableSourceClause / tableSource / tableQualifiedName / sourceReference / sideAlias). Legitimate column-minting `AS` (stats `... AS mean`, rename target, eval LHS) is unaffected. Adds a regression test; verified the test fails without the fix. Also two low-risk hardenings surfaced in review: - offendingFollowsPipe: bound tokenIndex by stream.size so a recovery- synthesized out-of-range index can't drive an out-of-bounds lookback. - edit_distance: document the three-row rolling rotation invariant. Signed-off-by: Hanyu Wei <weihanyu@amazon.com> * fix(ppl-lint): apply quick-fix even after the model version changes The lint/command-typo quick-fix lightbulb did nothing when clicked. The code-action provider stamped each workspace edit with the model's versionId captured at the moment the action was computed. Monaco's bulk-edit service rejects an edit whose versionId no longer matches the model ("bad state - model changed in the meantime") and the standalone editor's notification service swallows the error — so the button was a silent no-op. In the live editor the model version advances between computing the action and the user clicking it: the debounced re-lint, re-tokenization, and autocomplete all bump it. So the stale versionId virtually always mismatched by click time. It never reproduced in the jest unit tests because nothing there churns the version. Omit versionId (pass undefined), which skips the guard. The fix range is absolute, so applying without the version check is safe. Verified in a browser against the built bundle: emulating Monaco's exact bulk-edit guard, a pinned-version edit is rejected after version churn while the undefined-version edit applies. Adds a regression test asserting the provider never pins a versionId. Signed-off-by: Hanyu Wei <weihanyu@amazon.com> * fix(ppl-lint): load Monaco codeAction contribution so quick-fixes apply The lint/command-typo quick-fix lightbulb rendered nothing and clicking it was a silent no-op. osd-monaco's `monaco.ts` imports individual Monaco editor contributions (suggestController, hoverContribution, parameterHints, formatActions, comment) but never the codeAction one. So `monaco.languages.registerCodeActionProvider` stored the PPL provider (the fix even showed up in the side-table registry, and hover worked) while the module that renders the lightbulb widget and registers the `editor.action.quickFix` command was never loaded — there was no UI to invoke the provider. Import `contrib/codeAction/browser/codeActionContributions` (the same module the canonical `editor.all` bundle loads; it registers CodeActionController, LightBulbWidget, and QuickFixAction). Browser-verified against the built bundle: pre-fix the editor threw "command 'editor.action.quickFix' not found" with no widget; after the fix the lightbulb renders and clicking "Replace with 'balance'" rewrites `balnce` -> `balance` in place. Signed-off-by: Hanyu Wei <weihanyu@amazon.com> * fix(ppl-lint): generalize field-slot message and default field-validation to error Two related field-validation adjustments: - The field-slot shape message hardcoded "PPL does not use Splunk-style 'field=' syntax", but the pass fires on ANY non-bare-field shape — a bare comparison (`grok status > 200`), a literal, or arithmetic — not only `field=`. The Splunk framing misdescribes those cases. Drop it; the message is now just "<cmd> expects a field name here, not an expression." A test asserts the message no longer mentions Splunk. - Default field-validation severity warning -> error in both the catalog and the query_enhancements per-rule defaults (kept in lock-step). An unknown field or a Splunk-style field slot is not a stylistic heads-up: the query will fail at the engine, so error is the right default. Still user-toggleable via the per-rule uiSetting; the pass continues to honor config.severity. Browser-verified on the live runtime-grammar surface: `grok field=url` and `grok status > 200` both surface the generic message at severity Error, and `where balnce > 1` now reports Error with the "Did you mean balance?" fix. Signed-off-by: Hanyu Wei <weihanyu@amazon.com> * docs(ppl-lint): explain shouldUseRuntimeGrammar's optimistic version gate `shouldUseRuntimeGrammar` returns true for an unknown version while `shouldFetchFromBackend` returns false — which reads like a bug but is deliberate. The former is a synchronous render-time gate that cannot resolve the version (local needs an async /api/status call, remote an async saved-object read), so it optimistically prefers runtime and lets the async warmUp -> resolveVersion -> shouldFetchFromBackend chain make the real decision. If the grammar can't load, the runtime lint/validate paths return null and the editor falls back to the compiled grammar. Add a comment so this isn't "fixed" to return false on unknown version, which would break the local cluster whose version is only knowable asynchronously. Signed-off-by: Hanyu Wei <weihanyu@amazon.com> * feat(ppl-lint): make the command-typo suggestion toggleable via uiSetting The command-typo suggestion ("Unknown command 'wherre'. Did you mean 'where'?" + quick-fix) had no user control, unlike the lint rules. It runs on the syntax channel (it rewrites ANTLR's raw parse error), not as a tree-walking lint rule, so it isn't in the lint catalog. Add a `command-suggestion` entry to the existing "PPL lint rules" uiSetting so it sits alongside the rules, plumbed through the syntax path rather than the catalog. - ui_settings: add `{ id: 'command-suggestion', enabled: true }` to the defaults and make `severity` optional in the schema (this toggle carries no severity — it inherits the syntax error's). - lint_overrides: `isCommandSuggestionEnabled(uiSettings)` reads the flag separately from the catalog-iterating override builder. Defaults to enabled; only an explicit `enabled: false` turns it off. - lint_context_builder: carry `commandSuggestionEnabled` on the lint context. - language.ts (syntax marker builder): when disabled, revert an UNKNOWN_COMMAND error to ANTLR's raw message and drop the quick-fix. To make "off" the true original behavior, the error listeners now preserve ANTLR's `rawMessage` alongside the friendly rewrite (both the compiled PPLSyntaxErrorListener and the runtime PPLCommandErrorListener), threaded through PPLValidationResult. Gating on the client (where both the compiled-worker and runtime validation paths converge) avoids plumbing the flag into the Web Worker. Unit tests cover the setting schema/defaults, the reader, the context wiring, and the syntax marker gate; browser-verified on the live editor: enabled -> "Did you mean 'where'?" + fix; disabled -> "mismatched input 'wherre'..." with no fix. Signed-off-by: Hanyu Wei <weihanyu@amazon.com> * chore(ppl-lint): retrigger CI Signed-off-by: Hanyu Wei <weihanyu@amazon.com> * fix(ppl-lint): gate command suggestion on global lint flag and key field cache on data source Address two review findings from Eric on PR opensearch-project#12298: 1. The command-typo suggestion enhancement (friendly message rewrite + quick-fix lightbulb) was not gated on the global `isPPLLintEnabled()` flag. When lint was disabled via `queryEnhancements.ppl.lint.enabled`, the suggestion still fired because the per-context toggle defaulted to true when no context was pushed. Now the syntax path checks isPPLLintEnabled() first; when off, users see the raw ANTLR error with no enhancement. 2. The field-validation cache matched only on dataset id. If two different data sources happened to share a dataset id, stale fields could be reused. Now the cache stamps and checks `dataSourceId` alongside `datasetId` in both the data plugin query_editor and the explore use_query_panel_editor hooks. Signed-off-by: Hanyu Wei <hanyu@amazon.com> Signed-off-by: Hanyu Wei <weihanyu@amazon.com> * chore(ppl-lint): retrigger CI Signed-off-by: Hanyu Wei <weihanyu@amazon.com> * feat(ppl-lint): fire field-validation shape pass on the compiled surface field-validation's field-slot shape pass (grok/parse/patterns field=body) was runtime-only: it hard-returned [] on the compiled-simplified grammar, where `field=body` error-recovers and can't be read off the parse tree. The existence pass already fired there, but the shape pass did not, so sub-3.6 clusters got at most a raw syntax squiggle instead of an actionable finding. Add a narrow, comment/quote-aware text-side scanner (findCompiledFieldSlotShapeMatches) that recognizes exactly the one backend-accepted `grok|parse|patterns field=<bareField>` typo from the raw query, and route the compiled surface to it via a new sourceText field on LintRunContext (set by PPLLanguageAnalyzer.lint). The compiled emit honors config.severity (catalog default error) so the rule keeps a single severity toggle across both grammar surfaces. Signed-off-by: Hanyu Wei <weihanyu@amazon.com> * chore(ppl-lint): reformat PPL files for prettier 3.9.4 The main merge (opensearch-project#12340) upgraded prettier 2.1.1 -> 3.9.4 and eslint 10.6.0, but the branch-new PPL files pre-date that reformat, so `yarn lint` failed on 16 prettier/prettier errors in six of them. Apply the prettier-3 formatting (strip redundant `(x as unknown)` cast parens, wrap mixed `??`, nested-ternary indentation, multiline type-cast layout). Formatting only, no behavior change. Signed-off-by: Hanyu Wei <weihanyu@amazon.com> * fix(ppl-lint): tighten pattern-field regex and validate override severity Backport two correctness fixes from poc-ppl-linter-v3 (1d1fd25). 1. extractCreatedFieldNames used a greedy `(?<([^>]*)>` opener that consumed across a preceding lookbehind and never matched the real named group, so a field a parse/rex/grok pattern creates (e.g. `(?<=user_)(?<username>\w+)`) was dropped from field-validation's allowlist and then falsely flagged as an unknown field. Tighten the name charset to `[A-Za-z][A-Za-z0-9]*` so the regex cannot start on a `(?<=` / `(?<!` opener and never skips a real group. 2. buildOverridesFromSettings applied any stored severity string. An unknown value (reachable via the raw uiSettings API) makes `SEV_RANK[...]` undefined, so the MIN_SEVERITY floor comparison is false and the junk value slipped past the clamp that protects error-level rules. Ignore severities that are not real levels. Adds regression coverage for both (each new test fails on the pre-fix code and passes after). Signed-off-by: Hanyu Wei <weihanyu@amazon.com> * fix(ppl-lint): normalize backtick-quoted created field names field-validation false-flagged backtick-quoted created/derived fields as "Unknown field": references were normalized per dotted segment (unquoteIdent) but the created-field registration stored the raw getText() including backticks, so a valid query like source=t | eval `total` = age * 2 | where `total` > 100 flagged `total`. Five identifier slots had this asymmetry: stats/rename/etc. AS aliases, eval LHS, spath OUTPUT, the spath path fallback, and join side aliases (`left=`l``). Route both sides through a single shared normalizeFieldName helper (exported from pipeline_shape) so the created-field and reference sides can never drift. The helper strips one enclosing quote pair per dotted segment; single/double quotes are stripped too (created names only — e.g. `rename age as 'years'`), which is one-directional since references can only be backtick-quoted. Also correct stale docs: the field-validation docstring and a test mock said severity `warning`, but the catalog default was deliberately set to `error` in an earlier commit. No behavior change — catalog and uiSettings defaults are untouched; the rule stays user-downgradable via its per-rule uiSetting. Adds regression coverage for both reported repros plus rename/stats-alias, join-alias, and spath OUTPUT backtick cases, a mixed-quoting case, a negative control, and suggestion hygiene. Signed-off-by: Hanyu Wei <weihanyu@amazon.com> * fix(ppl-lint): sync doc links with structural rules Signed-off-by: Hanyu Wei <weihanyu@amazon.com> * chore(ppl-lint): retrigger CI after rebase Signed-off-by: Hanyu Wei <weihanyu@amazon.com> --------- Signed-off-by: Hanyu Wei <weihanyu@amazon.com> Signed-off-by: Hanyu Wei <hanyu@amazon.com> Co-authored-by: Hanyu Wei <weihanyu@amazon.com>
…rch (opensearch-project#12408) Make getTimeFilterWhereClause engine-aware: emit bare string literals for OpenSearch (which fold to native Lucene range queries) and only wrap in TIMESTAMP('...') for legacy Elasticsearch/Open Distro engines that reject bare strings. On OpenSearch, `field >= TIMESTAMP('...')` cannot be folded to a native range query because the RHS is a function expression, so it is pushed down as a per-value `opensearch_query_expression` script filter. Each unique timestamp compiles a new script and quickly exhausts the cluster-wide `script.context.filter.max_compilations_rate` limit (default 75/5m), producing `CircuitBreakingException` errors in classic Discover. Emitting a bare string lets the optimizer produce a native range query with zero script compilations, while Elasticsearch keeps the TIMESTAMP() wrapper it requires. Signed-off-by: Suchit Sahoo <suchsah@amazon.com>
Signed-off-by: Suchit Sahoo <suchsah@amazon.com>
…arch-project#12410) * [Explore] Warn when saving a complex query as a visualization Adds an `explore.queryProfiling.enabled` feature flag (default off). When it is enabled, the Explore save windows -- the top-nav Save modal, the Add to dashboard modal, and the in-context Save-as-visualization modal -- show a warning banner telling the user the query is complex: it can use significant cluster resources and re-runs on every dashboard load and refresh. The banner is informational and the Save / Add action is unchanged. The banner reuses each modal's existing EuiCallOut pattern and the shared SavedObjectSaveModal `options` slot, so the save flow itself is untouched. The per-query "complex" signal will come from query profiling; until that is wired the banner is shown whenever the flag is enabled (see the TODOs at the call sites). Signed-off-by: Jialiang Liang <ryanleeang@gmail.com> * Add changelog fragment for opensearch-project#12410 Signed-off-by: Jialiang Liang <ryanleeang@gmail.com> * [Explore] Gate the complex-query warning on query profiling Wires the complex-query warning banner to the real per-query signal from query profiling, replacing the feature-flag placeholder. - When `explore.queryProfiling.enabled` is on, Explore sends `profile: true` on the PPL query (query_actions.ts -> data plugin Query.profile -> query_enhancements facet.ts). - The PPL search strategy reads the backend's `profile.thread_pool` from the response body and surfaces `isComplex` (true for the `sql-complex-worker` pool) on the data frame meta. - executeQueryBase captures `isComplex` into the results slice; the three save windows gate the banner on the stored value for the query being saved (via useIsQueryComplex, and the store for the non-component top-nav Save handler). Signed-off-by: Jialiang Liang <ryanleeang@gmail.com> * [Explore] Allow the profile flag in the query-enhancements search route schema The search request body validated `query` strictly, so the `profile` flag added for query profiling was rejected ("definition for this key is missing") on every query. Allow it as an optional boolean. Signed-off-by: Jialiang Liang <ryanleeang@gmail.com> * [Explore] Show the complex-query warning in the in-context visualization editor The in-context editor (create/edit a visualization from within a dashboard) runs on its own query-builder state with no Redux Provider, so it reads the complex signal from its own result observable rather than the Explore results slice: - Capture `isComplex` from the data frame meta into the editor's query result (query_builder/utils.ts), mirroring executeQueryBase. - SaveVisButton reads `resultState.isComplex` via useQueryBuilderState and shows the banner in its save modal. Signed-off-by: Jialiang Liang <ryanleeang@gmail.com> * [Explore] Guard complex-query detection against unpopulated query state The save-window complex-query warning reads the Redux store and query slice to decide whether to show the banner. When the store or query state is not yet populated, these reads threw and broke the save button (surfaced by the existing add_to_dashboard_button and top_nav_save unit tests). The banner is a non-critical enhancement, so guard both read sites to degrade to 'no banner' instead of throwing. Signed-off-by: Jialiang Liang <ryanleeang@gmail.com> * [Explore] Reword complex-query warning to drop 'cluster' Reword the save-window warning banner from 'significant cluster resources' to 'resource-intensive' so the message reads clearly for users regardless of deployment terminology. Signed-off-by: Jialiang Liang <ryanleeang@gmail.com> * [Explore] Add runtime advanced setting to toggle query profiling Add an 'Enable query profiling' advanced setting (explore:enableQueryProfiling, default off, category explore) so admins can turn the complex-query warning on/off at runtime from /app/settings without a config change. The setting is registered by the Explore plugin, so it only appears when Explore is enabled. Profiling now requires BOTH the deployment-level config gate (explore.queryProfiling.enabled) AND the advanced setting: the outgoing profile:true flag is sent only when both are on. The setting is read live at query execution time so the toggle takes effect on the next query. When the config gate is off the setting has no effect, keeping the feature off by default. Adds unit tests for the setting registration and the combined gate. Signed-off-by: Jialiang Liang <ryanleeang@gmail.com> * [Explore] Remove the query profiling advanced setting; keep the config flag Drop the explore:enableQueryProfiling advanced setting and its runtime gate, reverting to a single deployment-level switch: the explore.queryProfiling.enabled config flag. Operators who want to drive it from the UI can use the native uiSettings.overrides mechanism. Also removes the now-redundant comment on Query.profile. Signed-off-by: Jialiang Liang <ryanleeang@gmail.com> * [Explore] Limit query profiling to PPL queries createSearchSourceWithQuery is shared across languages, so gating the profile flag on the config flag alone also sent it on SQL/PROMQL queries. Only PPL runs on the complex worker pool the profile response reports on, and the extra field is meaningless (and can affect engine selection on some backends) for other languages. Gate on preparedQuery.language === 'PPL' as well as the flag. Addresses review feedback on opensearch-project#12410. Signed-off-by: Jialiang Liang <ryanleeang@gmail.com> * [Explore] Nest query profiling result under a profile object Group the profiling signal under a `profile` object (QueryProfile) on the data frame meta and the search result, instead of exposing `isComplex` as a top-level attribute, so future profiling fields stay nested together. Readers now go through `profile?.isComplex`. Addresses review feedback on opensearch-project#12410. Signed-off-by: Jialiang Liang <ryanleeang@gmail.com> --------- Signed-off-by: Jialiang Liang <ryanleeang@gmail.com>
Signed-off-by: yubonluo <yubonluo@amazon.com>
opensearch-project#12421) * Change the allowed title length to 64 to comply with the collection title length Signed-off-by: Ella Zhu <zhyuanqi@amazon.com> * fix tests Signed-off-by: Ella Zhu <zhyuanqi@amazon.com> --------- Signed-off-by: Ella Zhu <zhyuanqi@amazon.com>
…1.7.6, diff 8.0.4 (opensearch-project#12406) * chore(deps): puppeteer 24.43.1, ws 8.21.1 Signed-off-by: Tomasz Kania <tomasz.kania@pl.ibm.com> * chore(deps): mocha 11.7.6 Signed-off-by: Tomasz Kania <tomasz.kania@pl.ibm.com> * chore(deps): body-parser 2.3.0 Signed-off-by: Tomasz Kania <tomasz.kania@pl.ibm.com> * chore(deps): diff 8.0.4 Signed-off-by: Tomasz Kania <tomasz.kania@pl.ibm.com> * fix: ensures id never shorter than 9 chars as test expects Signed-off-by: Tomasz Kania <tomasz.kania@pl.ibm.com> --------- Signed-off-by: Tomasz Kania <tomasz.kania@pl.ibm.com>
- Bump @elastic/charts 31.1.0 -> 71.8.1 (root package.json and osd-ui-shared-deps). This also pulls uuid ^14, removing the vulnerable uuid 3.x and resolving CVE-2026-41907. - postinstall.js: patch charts dist/theme.scss for constructs that OUI 1.22.1 / Dart Sass reject: euiFocusRing(null, 1) (OUI's mixin takes a single arg) and legacy $euiSizeXS division. - Discover histogram: deep-clone the charts default theme before applying overrides, since the theme object is frozen in charts >=71 (fixes 'Cannot assign to read only property axisTitle' crash), and set timeAxisLayerCount=0 to keep the single-row time axis so the existing tickFormat drives x-axis labels. Signed-off-by: Justin Kim <jungkm@amazon.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
20 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Upgrades
@elastic/chartsfrom 31.1.0 → 71.8.1.Changes
package.json,packages/osd-ui-shared-deps/package.json): 31.1.0 → 71.8.1. This pullsuuid ^14, removing the vulnerableuuid 3.xand resolving CVE-2026-41907.scripts/postinstall.js: patch chartsdist/theme.scssfor constructs incompatible with OUI 1.22.1 / Dart Sass:euiFocusRing(null, 1)→euiFocusRing(OUI's mixin takes a single arg)$euiSizeXSSass division incalc()histogram.tsx):Cannot assign to read only property 'axisTitle'timeAxisLayerCount={0}to keep the single-row time axis so the existingtickFormatcontinues to drive x-axis labels (charts >=71 defaults to a multi-layer ordinal time axis)Testing
yarn osd bootstrapsucceeds; all 61 dev bundles compile.Note: charts >=71 uses an adaptive, width-aware tick algorithm, so the histogram shows fewer x-axis labels than 31.1.0 at the same width (format is unchanged).