Repository navigation
Resource Governor
Under Management: resource pools with their workload groups nested beneath, and external resource pools. Pools and groups show live counters (active and queued requests, CPU used) beside their limits, and the node's label says whether the governor is disabled or has a reconfiguration pending.

- Resource Governor — the governor itself: its state and the live counters.
- Resource Pools → each pool → Workload Groups under it.
- External Resource Pools — the pools an external (PolyBase / SQL Server Big Data / Azure) request runs in, on the editions that have them.
Rows the edition does not support, and objects the login cannot see, each get their own row naming the reason rather than an empty folder.
The right it takes is asymmetric, and the pages say so: VIEW ANY DEFINITION to
read the configuration at all, CONTROL SERVER to change it. Read it and you get
the pages; not quite, and what you get is the sentence telling you which is missing.
Configuration
| Enabled | The governor's state. Clear it and Apply ends with ALTER RESOURCE GOVERNOR DISABLE — the stored changes stay stored, and a disabled governor is kept disabled rather than being silently reconfigured into being enabled. |
| Classifier function | The function that decides which group a request lands in. The list holds the schema-bound, parameterless functions in master that return sysname, with New to start one from a template. |
| Max outstanding I/O per volume | The I/O resource cap; 0 lets the server choose. |
In force — what sys.resource_governor_configuration says is actually running,
beside what has been written but not yet put in force. This is the pair the tree's
"(reconfiguration pending)" label comes from.
OK and Apply end with ALTER RESOURCE GOVERNOR RECONFIGURE, which is what puts
every stored change in force. That single trailing statement is why the dialog is worth
using instead of typing the CREATE RESOURCE POOLs by hand: a stored-but-not-in-force
configuration is one of the easy ways to lose an afternoon.
Affinity — the pool's NUMA-node and CPU affinity as the server holds it. It is
displayed, not edited: affinity is set with ALTER RESOURCE GOVERNOR … (AFFINITY …), and a page that listed editable CPU masks alongside it would be an
invitation to write something the server refuses.
Each pool page lists its workload groups beside the pool's own limits, with the live
counters — active requests, queued requests, CPU used — so a limit that is saturated is
visible as a limit that is saturated rather than a number out of context.
internal accepts no changes, and internal and default cannot be removed; a group
added on the page is added to the pool listed above it and is offered for selection at
once.
Each workload group page edits what CREATE WORKLOAD GROUP/ALTER WORKLOAD GROUP
takes, and writes only what changed:
| Importance | The group's request priority class |
| Max memory grant % | Request max memory grant percentage |
| Max CPU time per request | The per-request CPU cap, in the units the server uses |
| Memory grant timeout | How long a request waits for a grant before failing |
| Maximum DOP | The group's degree-of-parallelism cap |
| Maximum requests | The group's request ceiling |
| Max outstanding I/O per volume | The group's I/O cap; 0 lets the server choose |
| Tempdb data limit % / (MB) | tempdb space per request; blank means no limit |
| Uses external pool / Uses resource pool | For an external resource pool: which pool an external request uses |
0 means unlimited, or the server's own value, for CPU time, timeout, DOP and
requests, and the pages say which of the two they mean rather than leaving 0 to be
interpreted.
Add or remove creates a group (named, in the pool you were looking at) or marks one for removal — pending until Apply, and scriptable instead of run like everything else in a properties dialog.
Enable, Disable, Reconfigure, Reset Statistics, Delete and Script as are on the tree items, each gated on the right it needs and each offering the statement before it runs. Reconfigure is the one to reach for when something else changed the configuration — a script, another session — and the tree's "(reconfiguration pending)" label is the tell that something is stored and not running.
- Activity Monitor — the Azure Instance tab shows the governor's fixed limits beside the instance's own resource history
- Server and database properties — the configured-versus-running pattern again
- Scripting
- Getting connected
- Browsing and editing
- Diagnosing and tuning
- Server features