util/mon: reduce monitor-related allocations

[mgartner]

util/mon: rename fields of MonitorName

Release note: None

util/uuid: add comment to uuid.Short

Release note: None

util/mon: add fields to MonitorName

The processorID, suffix, and i fields will be used in future
commits to reduce allocations related to MonitorNames.

Release note: None

util/mon: compile-time checks for size of BytesMonitor and BoundAccount

Release note: None

sql: use mon.MonitorName in more places

Using structured mon.MonitorName objects reduces allocations of
strings in some cases. This commit doesn't replace all usages of string
monitor names - there are myriad such usages. We can iteratively chip
away at them in future commits.

Release note: None

util/mon: reduce size of MonitorName

The size of MonitorName has been reduced by replacing the string
suffix with a suffix enum.

Release note: None

util/mon: further reduce the size of MonitorName

The size of MonitorName has been reduced by using the same 4 bytes to
store either the int32 processor ID or the uuid.Short, and by
reducing the optional integer suffix from an int32 to uint16, which
should be sufficient in all cases.

Release note: None

util/mon: rename MonitorName to Name

Release note: None

util/mon: remove deprecated NewMonitorWithStringName

All usages of the deprecated NewMonitorWithStringName function have been
replaced with NewMonitor, and the former has been removed.

Release note: None

util/mon: remove deprecated NewMonitorInheritWithLimitAndStringName

All usages of the deprecated NewMonitorInheritWithLimitAndStringName
function have been replaced with NewMonitorInheritWithLimit, and the
former has been removed.

Release note: None

colexec: batch allocations of BoundAccount

Allocations of BoundAccounts in createUnlimitedMemAccountsLocked are
now batched into a single allocation.

Release note: None

sql/rowflow: refactor monitor name

The monitor names created in (*rowBasedFlow).setupRouter now use
(*mon.Name).WithID to add the stream ID to the name instead of
allocating a new string.

Epic: None
Release note: None

[cockroach-teamcity]

This change is 

[mgartner]

It's a very minor reduction in allocations. But it is nice to have structured names instead of strings.

name                                              old allocs/op  new allocs/op  delta
FlowSetup/vectorize=true/distribute=true               162 ± 1%       160 ± 0%  -1.31%  (p=0.000 n=16+16)
FlowSetup/vectorize=true/distribute=false              138 ± 4%       136 ± 4%  -1.32%  (p=0.000 n=17+17)
FlowSetup/vectorize=false/distribute=true              160 ± 1%       160 ± 0%    ~     (p=0.455 n=17+16)
FlowSetup/vectorize=false/distribute=false             135 ± 0%       135 ± 0%    ~     (all equal)

[yuzefovich]

Nice cleanup!

Reviewed 2 of 2 files at r1, 1 of 1 files at r2, 2 of 2 files at r3, 1 of 1 files at r4, 33 of 33 files at r5, 8 of 8 files at r6, 3 of 3 files at r7, 77 of 77 files at r8, 17 of 17 files at r9, 14 of 14 files at r10, 1 of 1 files at r11, all commit messages.
Reviewable status: complete! 1 of 0 LGTMs obtained (waiting on @asg0451, @jbowens, @kev-cao, and @xinhaoz)

---

pkg/util/mon/bytes_usage.go line 281 at r8 (raw file):

}

// Name is used to identify monitors in logging messages. It consists of:

nit: let's also mention crdb_internal.node_memory_monitors virtual table where the monitor name shows up.

---

pkg/sql/rowflow/row_based_flow.go line 464 at r9 (raw file):

	diskMonitors := make([]*mon.BytesMonitor, len(spec.Streams))
	for i := range spec.Streams {
		// TODO(mgartner): Is a StreamID ever greater than math.MaxInt32? If

I don't think it can ever exceed max int32 in practice. StreamIDs are indexes into PhysicalInfrastructure.Streams slice, which contains all pair-wise "links" between two processors within the same plan, so I'd expect that we'd OOM-crash if stream ID reaches 1 million, way before max int 32.

[mgartner]

Reviewable status: complete! 0 of 0 LGTMs obtained (and 1 stale) (waiting on @asg0451, @jbowens, @kev-cao, @xinhaoz, and @yuzefovich)

---

pkg/sql/rowflow/row_based_flow.go line 464 at r9 (raw file):

Previously, yuzefovich (Yahor Yuzefovich) wrote…

I don't think it can ever exceed max int32 in practice. StreamIDs are indexes into PhysicalInfrastructure.Streams slice, which contains all pair-wise "links" between two processors within the same plan, so I'd expect that we'd OOM-crash if stream ID reaches 1 million, way before max int 32.

Makes sense, I'll add another commit to address this TODO then. Thanks!

---

pkg/util/mon/bytes_usage.go line 281 at r8 (raw file):

Previously, yuzefovich (Yahor Yuzefovich) wrote…

nit: let's also mention crdb_internal.node_memory_monitors virtual table where the monitor name shows up.

done.