FGCB_ADD_REMOVE is not a wait type: it is a latch class, and its waits show up as LATCH_EX and LATCH_SH. It protects a filegroup while files are added, dropped, grown or shrunk, so time on it is almost always file growth inside the workload. Run the check to read the latch class, then line it up with your autogrowth history.You will not find FGCB_ADD_REMOVE in sys.dm_os_wait_stats; it is not in Microsoft’s wait type list and does not exist as a wait on our SQL Server 2025 lab. It is a latch class in sys.dm_os_latch_stats, used to synchronize access to filegroups for add, drop, grow, and shrink file operations. FGCB is the File Group Control Block. Waits for it are counted as LATCH_EX or LATCH_SH, and the class is how you tell them apart.
Is It a Problem?
Yes, when it shows real time, because while one session grows a file, others that need that filegroup’s file information queue behind it. An operation waiting on a long autogrow holds its locks and latches for the whole growth. Occasional small amounts, a well-sized file growing once in a while, are not worth chasing; on our lab the class had 7 waits totalling 0.1 seconds.
The baseline to monitor is this class’s wait_time_ms between two readings, next to the number of growth events in the same window.
Common Causes
- Autogrowth during the working day, especially percentage growth on large files.
- Very small growth increments, such as 1 MB, on a file that many threads are allocating into at once, bulk loads being the usual example. Small increments mean many more growth events, and each one is another chance for threads to queue on this latch.
- Instant file initialization off, so data file growth has to zero the new space.
- Large log growths, which are zeroed except for growths of 64 MB or less from SQL Server 2022.
- tempdb sized too small at startup and growing under load.
- Normal allocation, in small amounts, because the latch is also taken in shared mode while the engine finds the next file in the filegroup to allocate from.
The Check To Run
Read the latch class; the view is cumulative since restart, so take two readings:
SELECT
latch_class,
waiting_requests_count,
wait_time_ms,
max_wait_time_ms
FROM sys.dm_os_latch_stats
WHERE latch_class = 'FGCB_ADD_REMOVE';
Then pull the growth events with Get Autogrowth History; if the long growths line up with the latch time, that is the cause. Long growths with instant file initialization off also show as PREEMPTIVE_OS_WRITEFILEGATHER. The query needs VIEW SERVER STATE.
What To Do
- Pre-size data and log files for their expected size plus headroom, so autogrowth is a backstop rather than routine.
- Turn on instant file initialization for data files.
- Use fixed growth sizes, not percentages, and keep log growths moderate; on SQL Server 2022 and later, log growths of 64 MB or less can skip zeroing.
- Size tempdb for its normal peak; it is recreated at its configured size on every restart.
The IO Family
These are I/O waits outside the page latch families: bulk file operations, file growth, backups and throttling. The main page is IO_COMPLETION; for reads and writes of data pages, see PAGEIOLATCH_SH.
- IO_COMPLETIONSynchronous I/O. Waiting for I/O to finish, generally I/O that is not for data pages.The main page for this family
- ASYNC_IO_COMPLETIONAsynchronous I/O. Bulk file work such as backups reading data files and checkpoints.Worth investigating
- WRITE_COMPLETIONWrites. A write operation in progress.Worth a look when it climbs
- FGCB_ADD_REMOVEFile growth. A latch class rather than a wait type, taken for file add, drop, grow and shrink.Worth investigating. This page.
- IO_QUEUE_LIMITI/O throttle. Documented as internal use only; an I/O governance limit.Worth investigating
- BAD_PAGE_PROCESSSuspect pages. The suspect page logger pacing itself; a corruption signal when present.Worth a look when it climbs
- DISKIO_SUSPENDSnapshot backup. Waiting to access a file while an external backup is active.Worth a look when it climbs
- REPLICA_WRITESSnapshot writes. Page writes to database snapshots or DBCC replicas.Worth a look when it climbs
- FCB_REPLICA_READ / WRITESnapshot file sync. Synchronised reads from, and page pushes into, a snapshot sparse file.Usually noise
- IMPPROV_IOWAITBulk load I/O. Waiting for a bulk load I/O to finish.Worth a look when it climbs
- ASYNC_DISKPOOL_LOCKFile creation. Synchronising parallel threads that create or initialise a file.Worth a look when it climbs
Frequently Asked Questions
Why does Get Wait Statistics not show FGCB_ADD_REMOVE?
Because it is not a wait type. Its time is inside LATCH_EX and LATCH_SH, and sys.dm_os_latch_stats is where the class is named.
Where To Go Next
If file growth is stalling your workload, these are the next places to look.
- Get Autogrowth History, when each file grew and how long it took.
- LATCH_EX, the latch class check, with FGCB_ADD_REMOVE among the classes.
- Instant File Initialization, the setting that makes data file growth fast.
- Performance & Troubleshooting, the scripts and guides for this part of the stack.
Leave a Reply