Feature/admin forth/1239/make bulk ai flow run unlimite #11

yaroslav8765 · 2026-02-11T14:51:59Z

No description provided.

…f data

Copilot

Pull request overview

This PR implements concurrent bulk AI flow processing with configurable record selection modes. It refactors the frontend state management from array-based tracking to Map-based storage, adds support for filtering-based record selection (in addition to manual checkbox selection), and implements concurrency control using the p-limit library.

Changes:

Added concurrencyLimit and recordSelector configuration options to control concurrent processing and record selection behavior
Refactored frontend state management from multiple synchronized arrays to a centralized Map-based approach with RecordState objects
Implemented per-record concurrent processing using p-limit instead of sequential batch processing
Added backend endpoints for fetching individual record data and filtered record IDs

Reviewed changes

Copilot reviewed 5 out of 6 changed files in this pull request and generated 14 comments.

Show a summary per file

File	Description
types.ts	Adds new plugin options: concurrencyLimit (default 10) and recordSelector ('checkbox' or 'filtered')
index.ts	Adds get_old_data and get_filtered_ids endpoints; fixes recordId validation logic
custom/package.json	Adds p-limit dependency for concurrency control
custom/package-lock.json	Lockfile update for p-limit and yocto-queue dependencies
custom/VisionTable.vue	Refactors to use record objects from parent instead of managing separate arrays; simplifies prop interface
custom/VisionAction.vue	Major refactoring to Map-based state management with per-record concurrent processing; implements filtered record selection

Files not reviewed (1)

custom/package-lock.json: Language not supported

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

Copilot · 2026-02-11T14:58:45Z

custom/VisionAction.vue

+function getOrCreateRecord(recordId: string | number): RecordState {
+  const key = String(recordId);
+  let record = recordsById.get(key);
+  if (!record) {
+    record = createEmptyRecord(recordId);
+    record.isChecked = !uncheckedRecordIds.has(key);
+    recordsById.set(key, record);
  }
+  return record;


The recordsById Map uses string keys (via String(recordId)), but record IDs from the backend could be numbers. While this normalization is consistent throughout the code, there's a risk if the backend returns mixed types or if string-based IDs contain leading zeros which would be lost in number conversion. The code at line 377 checks !record.isChecked but this flag can become stale if users interact with checkboxes during processing. Consider adding validation to ensure ID type consistency or document the expected ID type.

Copilot · 2026-02-11T14:58:45Z

custom/VisionAction.vue

+    const saveResults = await Promise.all(saveTasks);
+    const failedResult = saveResults.find(res => res?.ok === false || res?.error);
+
+    if (!failedResult) {
      confirmDialog.value.close();
      props.updateList();
-      props.clearCheckboxes();
-    } else if (res.ok === false) {
+      props.clearCheckboxes?.();
+    } else if (failedResult.ok === false) {
      adminforth.alert({
-        message: res.error,
+        message: failedResult.error,
        variant: 'danger',
        timeout: 'unlimited',
      });
      isError.value = true;
      errorMessage.value = t(`Failed to save data. You are not allowed to save.`);
    } else {
-      console.error('Error saving data:', res);
+      console.error('Error saving data:', failedResult);
      isError.value = true;
      errorMessage.value = t(`Failed to save data. Please, try to re-run the action.`);
    }


When saving records concurrently, if any save operation fails, only the first failed result is reported to the user via adminforth.alert. Other failed records are silently ignored. Users won't know which records failed to save if there are multiple failures. Consider collecting all failures and reporting them together, or implementing retry logic for failed saves, so users can identify which records need to be re-processed.

Copilot · 2026-02-11T14:58:45Z

custom/VisionAction.vue

 async function regenerateCell(recordInfo: any) {
  if (coreStore.isInternetError) {
    adminforth.alert({
      message: t('Cannot regenerate column while internet connection is lost. Please check your connection and try again.'),
      variant: 'danger',
      timeout: 'unlimited',
    });
    return;
  }
-  if (!regeneratingFieldsStatus.value[recordInfo.recordId]) {
-    regeneratingFieldsStatus.value[recordInfo.recordId] = {};
+  const record = recordsById.get(String(recordInfo.recordId));
+  if (!record) {
+    return;
+  }
+  if (!record.regeneratingFieldsStatus[recordInfo.fieldName]) {
+    record.regeneratingFieldsStatus[recordInfo.fieldName] = false;
  }
-  regeneratingFieldsStatus.value[recordInfo.recordId][recordInfo.fieldName] = true;
+  record.regeneratingFieldsStatus[recordInfo.fieldName] = true;
+  touchRecords();
  const actionType = props.meta.outputFieldsForAnalizeFromImages?.includes(recordInfo.fieldName)
    ? 'analyze'
    : props.meta.outputPlainFields?.includes(recordInfo.fieldName)
      ? 'analyze_no_images'
      : null;
  if (!actionType) {
    console.error(`Field ${recordInfo.fieldName} is not configured for analysis.`);
+    record.regeneratingFieldsStatus[recordInfo.fieldName] = false;
+    touchRecords();
    return;
  }

  let generationPromptsForField = {};
  if (actionType === 'analyze') {
    generationPromptsForField = generationPrompts.value.imageFieldsPrompts || {};
-    isAnalizingFields.value = true;
+    record.aiStatus.analyzedImages = false;
  } else if (actionType === 'analyze_no_images') {
    generationPromptsForField = generationPrompts.value.plainFieldsPrompts || {};
-    isAnalizingImages.value = true;
+    record.aiStatus.analyzedNoImages = false;
  }

  let jobId;
  let res;
  try {
    res = await callAdminForthApi({
      path: `/plugin/${props.meta.pluginInstanceId}/create-job`,
      method: 'POST',
      body: {
        fieldToRegenerate: recordInfo.fieldName,
        recordId: recordInfo.recordId,
        action: actionType,
        actionType: "regenerate_cell",
        prompt: generationPromptsForField[recordInfo.fieldName] || null,
      },
      silentError: true,
    });
  } catch (e) {
-    regeneratingFieldsStatus.value[recordInfo.recordId][recordInfo.fieldName] = false;
+    record.regeneratingFieldsStatus[recordInfo.fieldName] = false;
    console.error(`Error during cell regeneration for record ${recordInfo.recordId}, field ${recordInfo.fieldName}:`, e);
  }
  if ( res.ok === false) {
    adminforth.alert({
      message: res.error,
      variant: 'danger',
    });
    isError.value = true;
    errorMessage.value = t(`Failed to regenerate field`);
-    regeneratingFieldsStatus.value[recordInfo.recordId][recordInfo.fieldName] = false;
+    record.regeneratingFieldsStatus[recordInfo.fieldName] = false;
    return;
  }
  jobId = res.jobId;
  res = {}
  do {
    res = await callAdminForthApi({
      path: `/plugin/${props.meta.pluginInstanceId}/get-job-status`,
      method: 'POST',
      body: { jobId },
      silentError: true,
    });
    if (actionType === 'analyze') {
      await new Promise(resolve => setTimeout(resolve, props.meta.refreshRates?.fillFieldsFromImages));
    } else if (actionType === 'analyze_no_images') {
      await new Promise(resolve => setTimeout(resolve, props.meta.refreshRates?.fillPlainFields));
    }
  } while (res.job?.status === 'in_progress');
  if (res.job?.status === 'failed' || !res.ok || !res) {
    adminforth.alert({
      message: t(`Regeneration action failed for record: ${recordInfo.recordId}. Error: ${res.job?.error || 'Unknown error'}`),
      variant: 'danger',
      timeout: 'unlimited',
    });
    if (actionType === 'analyze') {
-      imageToTextErrorMessages.value[recordInfo.recordId][recordInfo.fieldName] = res.job?.error || 'Unknown error';
-      isAnalizingFields.value = false;
+      record.imageToTextErrorMessages[recordInfo.fieldName] = res.job?.error || 'Unknown error';
+      record.aiStatus.analyzedImages = true;
    } else if (actionType === 'analyze_no_images') {
-      textToTextErrorMessages.value[recordInfo.recordId][recordInfo.fieldName] = res.job?.error || 'Unknown error';
-      isAnalizingImages.value = false;
+      record.textToTextErrorMessages[recordInfo.fieldName] = res.job?.error || 'Unknown error';
+      record.aiStatus.analyzedNoImages = true;
    }
-    regeneratingFieldsStatus.value[recordInfo.recordId][recordInfo.fieldName] = false;
+    record.regeneratingFieldsStatus[recordInfo.fieldName] = false;
+    touchRecords();
    return;
  } else if (res.job?.status === 'completed') {
-    const index = selected.value.findIndex(item => String(item[primaryKey]) === String(recordInfo.recordId));
-
-    const pk = selected.value[index]?.[primaryKey];
-    if (pk) {
-      selected.value[index] = {
-        ...selected.value[index],
-        ...res.job.result,
-        isChecked: true,
-        [primaryKey]: pk,
-      };
-    }
+    record.data = {
+      ...record.data,
+      ...res.job.result,
+    };
    if (actionType === 'analyze') {
-      if (imageToTextErrorMessages.value[index]) {
-        imageToTextErrorMessages.value[index][recordInfo.fieldName] = '';
-      }
-      isAnalizingFields.value = false;
+      record.imageToTextErrorMessages[recordInfo.fieldName] = '';
+      record.aiStatus.analyzedImages = true;
    } else if (actionType === 'analyze_no_images') {
-      if (textToTextErrorMessages.value[index]) {
-        textToTextErrorMessages.value[index][recordInfo.fieldName] = '';
-      }
-      isAnalizingImages.value = false;
+      record.textToTextErrorMessages[recordInfo.fieldName] = '';
+      record.aiStatus.analyzedNoImages = true;
    }
-    regeneratingFieldsStatus.value[recordInfo.recordId][recordInfo.fieldName] = false;
+    record.regeneratingFieldsStatus[recordInfo.fieldName] = false;
+    touchRecords();
  }
 }


In the regenerateCell function, the polling logic doesn't use the reusable pollJob function and instead implements its own polling with a simple setTimeout. This creates code duplication and inconsistency - the polling logic differs from the main processing flow. The regenerateCell polling also lacks the checkIfDialogOpen() safety check that pollJob has, potentially continuing to poll even after the dialog is closed. Consider refactoring to use the shared pollJob function or at minimum add dialog-open checks.

Copilot · 2026-02-11T14:58:45Z

types.ts

+  concurrencyLimit?: number;
+
+  /**
+   * Defines the way how records are selected for the action. 


There's a trailing space after the period on line 129 which creates inconsistent spacing in the JSDoc comment.

Suggested change

* Defines the way how records are selected for the action.

* Defines the way how records are selected for the action.

index.ts

Copilot · 2026-02-11T14:58:47Z

custom/VisionTable.vue

      <template #cell:checkboxes="{ item }">
        <div class="max-w-[100px] flex items-center justify-center">
          <Checkbox
-            v-model="selected[tableColumnsIndexes.findIndex(el => el[primaryKey] === item[primaryKey])].isChecked"
+            v-model="item.isChecked"
          />
        </div>
      </template>


When recordSelector is set to 'filtered', checkboxes are still displayed in the table and users can interact with them. However, there's no mechanism to synchronize checkbox state changes with the uncheckedRecordIds Set. According to the documentation, 'filtered' mode should apply actions to all records matching filters "without showing any checkboxes". Consider either hiding the checkbox column entirely when in 'filtered' mode, or if checkboxes should remain visible for manual deselection, implement proper synchronization between item.isChecked and uncheckedRecordIds.

Copilot · 2026-02-11T14:58:47Z

custom/VisionAction.vue

+async function runAiActions() {
  popupMode.value = 'generation';
-
-  if (props.meta.isImageGeneration) {
-    isGeneratingImages.value = true;
-    runAiAction({
-      endpoint: 'initial_image_generate',
-      actionType: 'generate_images',
-      responseFlag: isAiResponseReceivedImage,
-    });
-  }
-  if (props.meta.isFieldsForAnalizeFromImages) {
-    isAnalizingImages.value = true;
-    runAiAction({
-      endpoint: 'analyze',
-      actionType: 'analyze',
-      responseFlag: isAiResponseReceivedAnalizeImage,
-    });
-  }
-  if (props.meta.isFieldsForAnalizePlain) {
-    isAnalizingFields.value = true;
-    runAiAction({
-      endpoint: 'analyze_no_images',
-      actionType: 'analyze_no_images',
-      responseFlag: isAiResponseReceivedAnalizeNoImage,
-    });
+  if (!await checkRateLimits()) {
+    return;
  }
+  const limit = pLimit(props.meta.concurrencyLimit || 10);
+  const tasks = recordIds.value
+    .filter(id => !uncheckedRecordIds.has(String(id)))
+    .map(id => limit(() => processOneRecord(String(id))));
+  await Promise.all(tasks);
 }


When the dialog is closed during active processing (via checkIfDialogOpen checks), concurrent operations are abandoned mid-flight but their associated jobs continue running on the backend. The job polling stops, but the backend jobs don't get cancelled. This can lead to wasted server resources and rate limit consumption for operations whose results will never be used. Consider adding a cleanup mechanism to cancel pending backend jobs when the dialog is closed, or at minimum document this behavior for operators.

Copilot · 2026-02-11T14:58:48Z

types.ts

+   * 'checkbox' means that user will select records manually by checkboxes, 
+   * 
+   * 'filtered' means that action will be applied to all records matching current 
+   *  filters without showing any checkboxes (use with caution). 


There's a trailing space after the period on line 134 which creates inconsistent spacing in the JSDoc comment.

Suggested change

* filters without showing any checkboxes (use with caution).

* filters without showing any checkboxes (use with caution).

Copilot · 2026-02-11T14:58:48Z

index.ts

          //return { error: "Missing action type" };
        }
-        else if (!recordId) {
+        else if (!recordId && typeof recordId !== 'number') {


The condition typeof recordId !== 'number' will always be true when combined with !recordId using AND operator. If recordId is 0 (a valid number), it would be falsy but pass the type check. This creates a logic error where recordId = 0 would incorrectly fail validation. Consider changing to (recordId === undefined || recordId === null) to properly handle numeric IDs including 0.

Suggested change

else if (!recordId && typeof recordId !== 'number') {

else if (recordId === undefined || recordId === null || typeof recordId !== 'number') {

Copilot · 2026-02-11T14:58:48Z

custom/VisionTable.vue

+const paginatedRecords = computed(() => props.records.slice(pagination.offset, pagination.offset + pagination.limit));
+
+const tableDataProvider = async ({ offset, limit }) => {
+  pagination.offset = offset;
+  pagination.limit = limit;
+  return {
+    data: paginatedRecords.value,
+    total: props.records.length,
+  };
+};


The pagination logic computes paginatedRecords by slicing the records array, but the tableDataProvider function is async and returns the already-computed paginatedRecords.value. This creates unnecessary complexity - the function doesn't need to be async since it performs no async operations. Additionally, the Table component likely expects the data provider to fetch data based on offset/limit, but here the entire records array is kept in memory and sliced. For large datasets (which is the point of unlimited bulk operations), this could cause memory issues. Consider whether server-side pagination would be more appropriate.

…ed checkbox

yaroslav8765 added 4 commits February 11, 2026 13:56

feat: refactor visionAction to use plimit for generation of big set o…

b8ef41d

…f data

feat: add ability to process records, which are filtered

ce467f9

fix: resolve performance problems by getting rid of huge reactive array

10b5d87

fix: stop generation when window is closed

23d4ebe

yaroslav8765 requested a review from Copilot February 11, 2026 14:52

Copilot started reviewing on behalf of yaroslav8765 February 11, 2026 14:52 View session

Copilot AI reviewed Feb 11, 2026

View reviewed changes

yaroslav8765 added 3 commits February 11, 2026 17:32

fix: don't use deprecated param "record" for the update

6185938

fix: block save button when generation in progress

e534a57

fix: fix checkboxes in table. Now it doesn't save record with uncheck…

b1e7985

…ed checkbox

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Feature/admin forth/1239/make bulk ai flow run unlimite #11

Feature/admin forth/1239/make bulk ai flow run unlimite #11

Uh oh!

yaroslav8765 commented Feb 11, 2026

Uh oh!

Copilot AI left a comment

Uh oh!

Copilot AI Feb 11, 2026

Uh oh!

Copilot AI Feb 11, 2026

Uh oh!

Copilot AI Feb 11, 2026

Uh oh!

Copilot AI Feb 11, 2026

Uh oh!

Uh oh!

Copilot AI Feb 11, 2026

Uh oh!

Copilot AI Feb 11, 2026

Uh oh!

Copilot AI Feb 11, 2026

Uh oh!

Copilot AI Feb 11, 2026

Uh oh!

Copilot AI Feb 11, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

	* Defines the way how records are selected for the action.
	* Defines the way how records are selected for the action.

	* filters without showing any checkboxes (use with caution).
	* filters without showing any checkboxes (use with caution).

	else if (!recordId && typeof recordId !== 'number') {
	else if (recordId === undefined \|\| recordId === null \|\| typeof recordId !== 'number') {

Feature/admin forth/1239/make bulk ai flow run unlimite #11

Are you sure you want to change the base?

Feature/admin forth/1239/make bulk ai flow run unlimite #11

Uh oh!

Conversation

yaroslav8765 commented Feb 11, 2026

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Reviewed changes

Uh oh!

Copilot AI Feb 11, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Feb 11, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Feb 11, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Feb 11, 2026

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Copilot AI Feb 11, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Feb 11, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Feb 11, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Feb 11, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Feb 11, 2026

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant