Quick answer
A Documentum xPlore index rebuild gets stuck or fails for one of five reasons: an interrupted online rebuild that left its in-progress marker in indexserverconfig.xml — which can also stop Dsearch from starting; a single document that kills the rebuild against Lucene’s 32,766-byte term limit; a schema change, such as facet sub-paths, that never got the online rebuild it requires; a full reindex quietly skipping corrupt or protected documents that nobody resubmits; or a live indexing pipeline that was the real problem all along. Start with the collection definition in indexserverconfig.xml and the rebuild log — and assume the client’s users will notice before your queue does.
WHAT YOU’RE SEEING
The rebuild ran in the client’s maintenance window. The ticket arrived anyway.
If you operate Documentum for clients, this one has a familiar shape. A migration, an upgrade, or a facet change called for an xPlore online rebuild or a full reindex. It ran in the window you negotiated. Days later, the client’s users report documents missing from search — or search is down entirely — while every infrastructure check on your board is green.
The signatures to look for: a rebuild log that prints Rebuilding index [dmftdoc] for collection [col1]. and then repeats Reindex for index col1.dmftdoc failed; a Dsearch that refuses to start, logging Failed to create index on library with a NullPointerException in ReindexStatus; or a rebuild that dies on the same document every attempt with Document contains at least one immense term. For scale expectations, community-documented runs put an online rebuild at hours to days and a full reindex of millions of documents at a couple of days — a small collection much past half a day is stuck, not slow.
One routing note before the fixes: if no rebuild was in play and search simply went stale — index agent stopped, queue backlog, transform failures — that’s the live pipeline, covered in our corporate guide: Documentum Search Not Returning Documents: Causes and Fixes.
CAUSES, RANKED BY LIKELIHOOD
Five ways a rebuild goes wrong
| # | Cause | How to confirm | Fix in short |
|---|---|---|---|
| 1 | Interrupted rebuild left its marker — the collection in indexserverconfig.xml still carries the in-progress rebuild property (community-reported: Build_dmftdoc), and Dsearch can fail startup on it | Inspect the collection block in indexserverconfig.xml; startup log shows Failed to create index on library, retries may add a port bind error | Back up the file, restore a clean collection definition, increment revision, restart, re-run the rebuild |
| 2 | A document kills the rebuild — content whose extracted token exceeds Lucene’s hard limit | java.lang.IllegalArgumentException: Document contains at least one immense term … bytes can be at most 32766 in length; same object ID fails every attempt | Remove or exclude the offending documents; worst case, recreate the collections empty and run a full reindex |
| 3 | Schema change without its rebuild — facet sub-paths added, but facets never appear | Facets missing in D2 after a “successful” full reindex | Run the online rebuild — ideally on the empty index before the full reindex, not after |
| 4 | The reindex quietly shed documents — corrupt, password-protected, or extraction-failing files skipped without stopping the run | IndexAgent logs show DM_INDEX_AGENT_RECEIVED_FT_CALLBACK_WARN, e.g. CPS Warning [Corrupt file]; Dsearch Admin undercounts | Audit the logs, resubmit the failed object IDs to the fulltext queue |
| 5 | It isn’t the rebuild — the live indexing pipeline went stale | No rebuild in play; pending queue climbing | See the corporate live-pipeline guide linked above |
THE FIXES, IN REPAIR ORDER
Work the chain from the config file outward
-
Clear a stuck rebuild marker
From a documented field incident (community-reported, one author’s verified recovery): back up
indexserverconfig.xml, replace the affected collection’s block with a clean definition — e.g.<collection usage="Data" document-category="dftxml" name="col1"/>— keeping any facet sub-paths, and increment the file’srevisionattribute so xDB reloads it on start (the log confirms with “File system configuration … is higher than xDB revision”). Restart Dsearch, then re-run the online rebuild so the index reflects current config. -
Get past a rebuild-killing document
Pull the failing object IDs from the rebuild log and remove or fix those documents — the documented case found enormous extracted content tripping Lucene’s 32,766-byte term limit, failing on the same document every retry. If documents keep surfacing, recreate cleanly: in the Dsearch Admin under
Home >> Data Management, delete thedftxml/Data collections, create them anew, run the online rebuild on the empty collections (near-instant), then launch the full reindex. OpenText’s KB15765485 describes a comparable cleanup. Plan the window: search returns nothing until the reindex completes. -
Make schema changes stick — rebuild before, not after
Facet and sub-path configuration is an index schema change, and every schema change requires at least an online rebuild — a full reindex alone keeps indexing on the old schema. The efficient order on a fresh or recreated index: apply the config, rebuild while the index is empty, then reindex once.
-
Audit what the reindex left behind
Errors don’t stop a reindex; they skip documents. The reliable record is the IndexAgent logs (
Indexagent_*.log), not the Dsearch Admin reports — in one documented 2.5-day reindex the logs surfaced error types the Admin UI never displayed, includingDM_STORAGE_E_BAD_TICKET. Grep for warning and error callbacks, collect the 16-character object IDs beginning09, and resubmit each via IAPI:queue,c,<id>,dm_fulltext_index_user.
That gets this client’s index back. The harder question is the one the fix can’t answer.
HOW TO STOP IT RECURRING — ACROSS EVERY CLIENT ESTATE
You don’t run one xPlore. You run one per client.
For an in-house administrator this is a bad week. For an MSP or SI it’s a structural exposure: every client estate has its own xPlore, its own rebuild history, its own quietly climbing failed-document count — and the first signal your desk usually gets is the client’s ticket. An index that drifts out of sync with the repository fails silently by design: the repository keeps accepting documents, search keeps answering queries, and nothing on an RMM dashboard distinguishes a complete index from one missing nine thousand documents.
Reveille SENTRY · MSPs · SIs · ISVs
Reveille ships more than 100 Documentum Content Management-specific tests, including an xPlore suite that watches exactly this failure chain: xPlore Index Agent Status, xPlore Index Pending Count, Index Failed Count, Index Warning Count, Index Processing Count, xPlore search instance available, and xPlore CPS instance available — plus search-path transaction tests (DFS Search Service – Query, CMIS Query) that prove a user’s search actually returns results. Through Reveille SENTRY, a partner and software program powered by Reveille, that watch runs as multi-tenant observability: one pane across every client estate, so a climbing Index Failed Count pages your NOC before it becomes the client’s discovery. Explore the SENTRY Platform →
Frequently asked questions
Why is my xPlore index rebuild stuck?
How long should an xPlore online rebuild take?
Why is Dsearch not starting after a rebuild?
Why are documents missing from Documentum search after a reindex?
The index drifts quietly. The watch doesn’t.
ONE PANE · EVERY CLIENT ESTATE · BEFORE THE TICKET




