Search configuration and index changes run asynchronously as background tasks. Confirm the outcome under History after submitting.
Query observation is off by default. Set
V2_TABLE_QUERY_OPS_ENABLED to true on the backend and restart to start collecting observations.Find the table to work on
The page opens on the table list. The state buttons across the top carry their own counts and narrow the list:
The search box accepts table, Project, and Space names or IDs. Sorting is either Most slow queries or Table name. Observation window covers the last 1 hour or 24 hours, and every observation figure in the list follows that window. Observations are kept for 2 days; anything older no longer appears in the list.
Each row gives the table’s estimated rows and size, the requests and slow queries within the window, the current search method and field coverage, the number of filter/sort indexes, and the counts of pending recommendations and active tasks. Click a table name to open its detail.
Use Choose a table at the top to switch tables directly. The selected table stays selected when you switch to Complexity or Schema integrity; Computed updates shows the queue for the whole instance.
Read the index state
The table list and the table detail both show an index state next to the search method:
Two lines can appear below the state. Configured: pg_bigm / pg_trgm means the saved provider differs from the search method in use, and Indexed search is switched off at runtime means the index is ready but the runtime is off, so search still runs
ILIKE.
When a notice at the top of the page reports tables whose configured search index is not serving, search on those tables falls back to ILIKE. If it also says the task worker is off, set V2_TABLE_QUERY_OPS_TASK_WORKER_ENABLED to true on the backend and restart so the indexes can be rebuilt automatically. You can also restore an index manually by choosing Restore search index on the table’s Search tab, which does not need the task worker. If the cause is that the indexed-search runtime switch is off, turn that switch on in the backend; restoring the index alone will not make it serve.
A published index does not mean every search hits it: a probe that is too short, or a field outside the coverage, still falls back to
ILIKE. This page reports the state of the index, not proof that recent requests hit it.Configure search coverage
The Search tab shows the table’s current search method, how many fields are covered, and the Serving coverage of each field. Show uncovered fields narrows the list to the fields not yet included; fields marked Not eligible cannot contribute to a substring search index.Serving coverage requires published, usable index metadata and the runtime switch on; a saved configuration alone does not count. Coverage is explicit, so fields you add later are not included automatically; choose Complete coverage to bring them in.
Teable picks an available search provider automatically; if the database lacks the
pg_bigm or pg_trgm extension, the dialog explains why. When restoring, the dialog opens with the fields saved last time; otherwise every eligible field is selected by default. Keep only the fields users actually search, and on a wide table clear the ones nobody searches; building a search document for every field is expensive.
Analyze and dry-run runs the change without committing it and reports the covered field count, estimated rows, whether plan evidence exists, and the impact of creating or rebuilding. Tables estimated at 50,000 rows or more, or of unknown size, also require the maintenance-window acknowledgement. Then choose Confirm and queue task.
What the dialog returns is a task receipt, not a result. Close it and follow the task under History. Submitting the same change again returns the existing task.
Manage filter and sort indexes
The Indexes tab lists filter/sort and system indexes, with their validity, management status, size and definition. Substring search indexes are listed separately under Search. Pending actions holds the system’s index recommendations and says whether plan evidence backs each one. Review index creation shows the proposed index, the database it runs against, and the evidence; after you confirm, Teable runsCREATE INDEX CONCURRENTLY asynchronously. Large tables and tables of unknown size need the maintenance-window acknowledgement here too.
This page does not delete indexes. Unmanaged and system indexes may protect constraints or serve other workloads, so they are left alone.
Diagnose specific queries
When someone reports a slow table and no recommendation covers it, open the Diagnosis tab and choose Analyze queries. The analysis only reads saved query shapes and available plan evidence; it creates nothing. The conclusion is one of Plan-backed index opportunities found, No index opportunity confirmed by the available plans, or Insufficient evidence. Insufficient evidence does not mean a sequential scan happened: a slow observation alone does not establish that the access path is at fault. Below it, each affected query is listed with whether it came from real traffic or a saved view configuration, and the estimated plan cost before and after the proposed index (plan cost, not milliseconds).Follow the results
The History tab holds three parts:- Tasks: background tasks for this table, with status, attempt count, task kind, and failure reason. When a task eventually succeeded after failing, the error is labelled as a prior attempt.
- Search configuration history: the provider, status, and field list of each past coverage configuration.
- Recommendation history: recommendations that were accepted, dismissed, or superseded.
While a table has a queued or running task, the buttons for maintaining the search index and creating indexes stay disabled. Wait for the task to finish. To cancel a queued or failed search task, choose Cancel unstarted task on the Search tab; a running task has to finish.
Assess change impact
Updating records can trigger further calculations in formulas, lookups and rollups through linked records. To assess whether a batch write, field conversion or field deletion could affect a large amount of data, open Complexity and select a table, or click Assess complexity in its search and index details. The Impact matrix estimates updates, field deletions and conversions for each field, plus record creation and deletion separately. The dot on the left shows the typical case; the fill shows the worst case. Click a cell for affected row estimates, propagation paths, limitations and recommendations. Review High or Critical changes first to decide whether to reduce batch sizes or adjust calculation dependencies. In Estimate settings, set Records per change to the number you plan to submit at once; leaving it empty estimates one record. Use catalog statistics is on by default and reads table sizes and link counts from the database to improve the estimate.Assessment reads structure and statistics only. It performs no record or field changes and does not predict elapsed time. If a result is Not estimated or its quality is Partial or Truncated, read the limitations and fill the information gaps before deciding. A missing estimate does not mean there is no risk.

