Index Management

It is important to index all assets and data object in Pimcore in order to be able to use the search and listing features powered by the Generic Data Index bundle.

For an overview of which changes require a reindex, which of them are handled automatically and which command to run for the rest, see the Reindexing Guide.

Define Index Prefix

To avoid duplicate names or index interferences in your search engine, it is necessary to define an index name prefix, which is added to all indices created by Generic Data Index. Default one is pimcore_.

This can be done by defining following configuration:

pimcore_generic_data_index:
    index_service:
        client_params:
            index_prefix: 'my_prefix' # Prefix for all index names created by Generic Data Index

Created Indices

The Generic Data Index generates indices for the following entities:

• Assets search index (one alias and one index)
• Data objects search index (one alias and one index per class definition)

For the asset and data object indices the Generic Data Index uses an alias (e.g. <index_prefix>_asset) that points to the most current index (e.g. <index_prefix>_asset-odd). The alias name always stays the same, the index names alternate between -odd and -even suffix. For more details also see 'Updating index structure for data indices' in the next section.

Why -odd and -even Suffixes?

The alternating suffixes enable zero-downtime reindexing (blue-green approach): when the index structure changes, the bundle builds a new index under the opposite suffix while the current index keeps serving reads and writes through the alias. Once the new index is fully populated, the alias is switched atomically and the old index is deleted.

At rest there is only one concrete index — and therefore one set of shards — per asset index or class definition. Both suffixes exist side by side only transiently while a reindex is running. If -odd and -even indices for the same alias exist permanently, they are leftovers from a reindex that was interrupted (e.g. by a killed process) before its cleanup could run. Such orphaned indices are not used by the bundle and can safely be deleted, as long as they are not referenced by any alias.

The default index settings use number_of_shards: 1, so the total shard count of the cluster scales with the number of class definitions — not with the suffix scheme.

Keeping Indices Up to Date

The element search indices need to be created via the following console commands:

# create/update all indices + their mappings and add all items to the index queue
bin/console generic-data-index:update:index

The command will create the indices and add all assets and data objects to the index queue. The queue will be processed by Symfony messenger workers (pimcore_generic_data_index_queue queue).

Refreshing of the index

By default, the index queue is refreshed after each bulk operation as the items are processed asynchronously with the Symfony messenger. If you want to perform index refresh immediately you can use enable synchronous processing by injecting the SynchronousProcessingServiceInterface and calling enable() method.

Available methods:

• enable(): enable synchronous processing
• disable(): disable synchronous processing
• isEnabled(): check if synchronous processing is enabled

Index Queue Options

The indexing queue considers the following options:

• worker_count (default 1): number of messenger workers to process the queue. Set this to the actual used parallel number of messenger:consume workers to improve the calculation of items per batch.
• min_batch_size (default 5): minimum number of items to process in one batch (when using multiple workers)
• max_batch_size (default 400): maximum number of items to process in one batch

Based on this configuration, the queue will be processed in batches of min_batch_size to max_batch_size items. The number of items per batch is calculated based on the number of workers and the number of items in the queue.

Sample configuration:

pimcore_generic_data_index:
    index_service:
        queue_settings:
            worker_count: 1
            min_batch_size: 5
            max_batch_size: 400

Related elements

The indexing queue is automatically populated whenever an element undergoes an update operation. This process includes not only the modified element itself but also any related elements. By default, this indexing occurs asynchronously through Symfony Messenger.

For scenarios requiring immediate processing, you can temporarily switch to synchronous mode by utilizing the SynchronousProcessingRelatedIdsServiceInterface.

Available methods are:

Method	Description
enable()	Activates synchronous processing mode
disable()	Reverts to asynchronous processing mode
isEnabled()	Returns the current processing mode status

info

Currently the SynchronousProcessingRelatedIdsServiceInterface interface does not influence the behavior of delete operations. They are always processed synchronously.

Repairing Indices

Sometimes it might be needed to delete and recreate the index from the Pimcore database (for example if the mapping changed and cannot be updated).

Do this with the index update command and pass -r option (which deletes and recreates the index).

# delete index and recreate it
bin/console generic-data-index:update:index -r

Without the -r option, the index mapping is just updated and all items are added into the queue for a reindex from the Pimcore database.

Updating Index Structure for Data Indices

Index mapping is updated automatically e.g. when adding system languages or new fields to the class definition. Sometimes it might be necessary to update the index structure manually.

Do this with the reindex command. This command does native opensearch/elasticsearch re-indexing. So it does not index data from the database but reindexes data within the search indices.

# updates index mapping with native reindexing
bin/console generic-data-index:reindex

Reindex Options

Native reindexing is submitted as an asynchronous task to OpenSearch/Elasticsearch and its progress is polled until the task has finished. Two options control how long the process waits:

• max_polls (default 720): maximum number of polling attempts before the reindex is aborted
• poll_interval (default 5): seconds to wait between two status polls

With the defaults, a reindex may therefore run for up to one hour (720 × 5 seconds). If reindexing an index takes longer than this budget, the command aborts with an error and a non-zero exit code — increase max_polls for very large indices:

pimcore_generic_data_index:
    index_service:
        reindex_settings:
            max_polls: 720
            poll_interval: 5

Transient failures while polling (e.g. a temporarily unreachable or overloaded search cluster) are retried and do not abort a running reindex. If reindexing fails for a different reason — the cluster is unreachable, the task reports an error or times out server-side — the command fails instead of modifying the live index, and the reindex is retried on the next run. The index is only recreated (and repopulated via the index queue) when the already indexed documents are incompatible with the new mapping, for example after a field type change.

Handling Failed Messages

By default, the messenger will retry failed messages 3 times and then send them into the failed queue pimcore_generic_data_index_failed. If you want to retry failed messages, you can use the following command:

php bin/console messenger:failed:retry -vv

For the further commands please refer to the Symfony Messenger documentation.

Index Status

An index doesn't show whether it is complete. After a mapping change the index is copied or recreated and then refilled from the database by the messenger workers; until they are done, filters, saved searches, exports and reports see incomplete data. The index status tells, per index, what state it is in and how far a rebuild has got, for PHP code, on the command line and as an event.

States

State	Meaning	Progress
complete	All elements are indexed. Queue entries still waiting are the normal lag of asynchronous indexing and are reported separately.	-
rebuilding	A full rebuild from the database was queued (after a class change, update:index, a recreated index) and its entries are not all processed yet.	elements already processed / elements; 0 while the elements are still being queued
degraded	Documents still carry the state of a copy that left out fields; the rebuild rewrites them. Only occurs after a copy that left out fields.	documents rewritten / documents
copying	An odd/even copy of the index is running in the search engine; search shows the old index until the alias switches.	-
incomplete	Documents are missing and nothing is queued: run update:index for the index.	-
missing	The index doesn't exist.	-

Each status also carries the counts it is derived from: documents in the index, elements in the database, waiting queue entries, documents without checksum, whether a copy runs, and whether the mapping and the index settings are current (null when no state is stored yet, see Deployment and Index Management). The measures are read live; nothing is cached.

A rebuild counts as queued from a full enqueue of the index (class save with a mapping change, update:index, deployment:reindex with a change, a recreated index): the mark is set before the first element is queued (workers draining the queue between two chunks of a large enqueue don't end the rebuild) and removed by exactly one process, the messenger worker that processes the last batch, or the enqueue itself when the workers were faster or the index has no elements. A mark left behind by a worker that died after its last batch doesn't hide missing documents: with an empty queue the counts decide, so such an index shows incomplete and asks for a rebuild. If the process queuing the elements was terminated or failed between two chunks, the mark keeps its queueing phase (the chunks are committed one by one, so part of the rebuild may be queued already): the index shows rebuilding at 0 % with rebuildEnqueuing and the time it was queued (the command prints rebuilding, queueing since …); run the full enqueue again (update:index, deployment:reindex), which replaces the mark. A rebuild queued by a version before this feature isn't marked, so the index shows complete with waiting entries, or incomplete once the queue is empty.

Command Line

bin/console generic-data-index:status                      # overview: queue, indices, status table
bin/console generic-data-index:status --index=Car          # one index: alias name, class name, asset, document
bin/console generic-data-index:status --json               # machine-readable
bin/console generic-data-index:status --check              # exit code 1 while an index is not complete
bin/console generic-data-index:status --wait --timeout=7200 --index=Car   # block until complete, exit code 1 on timeout

--wait and --check are meant for deployment pipelines: run the step that needs the index only after generic-data-index:status --wait has returned. --json leaves stdout to the report; errors, the failure of --check and the progress of --wait -v go to stderr. When the status cannot be read (search engine or database unavailable, the tasks API failing, a count or task list the engine answered only partially because of failed shards or nodes) the command exits with 1 instead of guessing, so an unknown state never passes as complete.

PHP

use Pimcore\Bundle\GenericDataIndexBundle\Service\SearchIndex\Status\IndexStatusServiceInterface;

public function __construct(private IndexStatusServiceInterface $indexStatusService)
{
}

public function export(ClassDefinition $classDefinition): void
{
    $status = $this->indexStatusService->getStatusForClassDefinition($classDefinition);
    if (!$status->isComplete()) {
        // $status->getState(), $status->getPercentage(), $status->getPendingEntries()
        throw new RuntimeException(sprintf('Index %s is %s', $status->getIndexName(), $status->getState()->value));
    }
    // ...
}

getAssetStatus(), getDocumentStatus(), getDataObjectFolderStatus() and getStatusByIndexName() read the other indices; getAll() returns a report with every index plus the total queue depth and the number of failed messages; waitUntilComplete($indexName, $timeoutSeconds) polls until the index is complete and throws an IndexStatusException on timeout.

Event

Pimcore\Bundle\GenericDataIndexBundle\Event\IndexStatus\IndexCompleteEvent fires once when the last queue entries of a queued full rebuild have been processed and the index is complete again. It is dispatched by the messenger worker that processes that batch, so listeners run in the worker; when the workers were faster than the enqueue, or the index has no elements, the process that queued the rebuild dispatches it. Two workers finishing their last batches at once dispatch one event, not two. If the search engine can't be reached right after the last batch, the check is repeated with the next batch of that index.

use Pimcore\Bundle\GenericDataIndexBundle\Event\IndexStatus\IndexCompleteEvent;
use Symfony\Component\EventDispatcher\Attribute\AsEventListener;

#[AsEventListener]
final class StartExportWhenIndexComplete
{
    public function __invoke(IndexCompleteEvent $event): void
    {
        $event->getIndexName();           // e.g. "pimcore_data-object_car"
        $event->getStatus()->getLabel();  // e.g. "class Car (CAR)"
    }
}

Configuring index options (Maximum Item Limit, ...)

You can configure different options to use with your indices. The available options can differ depending on which engine you are using. Make sure to check the corresponding documentation, before using any options.

See the Maxium Item Limit and Total fields limit section for examples.

Maximum Item limit

A maximum of 10000 items can be retrieved and viewed, because of the maximum default item limit. To increase this limit, configuration can be adjusted as follows:

pimcore_generic_data_index:
    index_service:
        index_settings:
            max_result_window: 20000

Total fields limit

A maximum of 1000 fields can be used with your indces. To increase this limit, configuration can be adjusted as follows:

pimcore_generic_data_index:
    index_service:
        index_settings:
            mapping.total_fields.limit: 20000

info

If an index was already created before setting this parameter, the index needs to be recreated.

bin/console generic-data-index:update:index -r

Deployment and Index Management

Pimcore Class Definitions

After every class definition update you should run the following command to update the index structure:

php bin/console generic-data-index:deployment:reindex

This command will update the index structure for all data object classes which were created/updated since the last deployment and reindex all data objects for relevant classes.

Other Index Changes

With index options the command also checks the data object folder, asset and document indices, and compares index_settings and data versions as well. This covers e.g. predefined asset metadata, valid languages from config files, system_fields_settings, search_analyzer_attributes and ExtractMappingEvent fields.

php bin/console generic-data-index:deployment:reindex --all

Option	Checked indices
--all	All of the below
--classes / --folders / --assets / --documents	Only the named indices
--dry-run	Prints the planned action per index, changes nothing

Detected change	Action
Mapping	Native reindex, then rebuild from the database
Index settings	Native reindex
Data version	Rebuild from the database
Index missing	Create the index, then rebuild from the database

The values are stored per index after a successful update; a failed index is retried on the next run. The first run with index options only stores the current values as the baseline. If that deployment also changes a mapping, run generic-data-index:update:index once; for index_settings, run generic-data-index:reindex once (it does not cover the data object folder index).

A native reindex switches the alias to a copy of the index; index updates processed during the copy are lost. Run the command while the index queue workers are stopped.

Code changes that alter indexed values without changing the mapping (calculators, UpdateIndexDataEvent subscribers, custom adapters) are not detected: raise the matching version in the same change. In query_store mode a rebuild reads the stored calculator values, see Refreshing calculated values after a calculator change.

pimcore_generic_data_index:
    index_versions:
        assets: 1        # also: documents, folders
        classes:
            CAR: 3       # class definition ID

Calculated Fields Index Mode

By default, values of calculated value fields are computed live while index data is extracted: the field's calculator class is executed for every calculated field of every element on every (re)indexing — for localized calculated fields once per language. For expensive calculators (relation loads, external services) this can dominate the indexing cost on large installations.

With the query_store mode, the value stored in the object's query table (written on every save — the same value SQL-based grid listings use) is indexed instead, and the calculator is never executed during indexing:

pimcore_generic_data_index:
    index_service:
        calculated_fields_index_mode: 'query_store' # default: 'live'

Notes:

• The indexed value is the save-time snapshot as a string, truncated to the field definition's columnLength (default 190). The fields stay searchable, filterable and sortable.
• Changed calculator logic is only reflected after an element is saved again (see Refreshing calculated values after a calculator change below).
• Calculated fields inside field collections and object bricks are not read from the query store and keep live behavior.
• Switching the mode does not change the index mapping, so it does not trigger an automatic reindex: values converge as elements are saved or reprocessed by the index queue.

Refreshing calculated values after a calculator change

In query_store mode the indexed value is the save-time snapshot stored in the object's query table (object_query_*), which is (re)written only when an object is saved. Changing a calculator's code therefore does not update already-stored values on its own — exactly as with Pimcore's classic SQL grids and listings, which read the same query-table snapshot.

To refresh calculated values after a calculator change, re-save the affected objects (e.g. a short console command that loads the class's objects in chunks and calls ->save() on each). A save recomputes the calculated fields, rewrites object_query_*, and enqueues the element for reindexing, so both the query table and the index pick up the new logic. For large object counts, batch the re-save and disable versioning for the run to reduce cost.

A dedicated command to recompute the query-table snapshots without a full object save is planned as a follow-up.