LATCH_UP Wait Type in SQL Server

---

LATCH_UP is a wait for an update-mode (UP) latch on an internal SQL Server structure that is not a data or index page. Update mode sits between shared and exclusive: the holder intends to modify the structure but still allows concurrent readers. The wait type exists so every latch mode has a matching LATCH_* entry; the mode itself tells you far less than the latch class involved.

For buffer pages the equivalent is PAGELATCH_UP; LATCH_UP is everything else.

Is It a Problem?

Rarely on its own, and never diagnosable from the mode alone. sys.dm_os_latch_stats groups all modes (NL, SH, UP, EX, DT) together per latch class, which is the correct altitude anyway: what matters is which internal structure is under contention, not which flavour of latch was requested.

If LATCH_UP registers meaningfully, some structure that gets modified-while-read is hot, and the class name will say which subsystem owns it.

Common Causes

• Contention on allocation and metadata structures under concurrent mixed read/modify access.
• The same workloads that drive LATCH_EX and LATCH_SH; the three usually travel together on one hot class.

What To Do

1. Query sys.dm_os_latch_stats ordered by wait time and take the top class as the real lead.
2. Follow the class to its subsystem, exactly as you would for LATCH_EX.
3. For before/after tests, clear latch stats with DBCC SQLPERF ('sys.dm_os_latch_stats', CLEAR); and measure just your window.

How To See It

Rank it against everything else with Get-WaitStatistics, then pivot to latch classes; the mode-level wait name has done its job once it has your attention.

---