Run this script:
#!/bin/bash
THREADS=$1
CMD=$2
if [ "$3" == "" ]; then
BENCH_SECONDS=10
else
BENCH_SECONDS="$3"
fi
mongo --quiet -eval 'db.foo.insert({_id:1})'
if [ $CMD == "false" ]; then
mongo --quiet -eval '
ops = [{op:"find", ns:"test.foo", query: {_id: 1}}]
results = benchRun({ops:ops, parallel:'$THREADS', seconds:'$BENCH_SECONDS'})
print(Math.round(results["totalOps/s"]))
'
else
mongo --quiet -eval '
ops = [{op:"command", command: {"find": "foo", "filter": {"_id" : 1}}, ns:"test"}]
results = benchRun({ops:ops, parallel:'$THREADS', seconds:'$BENCH_SECONDS'})
print(Math.round(results["totalOps/s"]))
'
fi
You get much more lock contention with the command path vs not:
benchRun.find 32 false 30
233511 ops/s
- 6.32% mongod [kernel.kallsyms] [k] _raw_spin_lock
- _raw_spin_lock
- 48.54% futex_wake
do_futex
sys_futex
system_call_fastpath
- __lll_unlock_wake
+ 41.15% mongo::PlanExecutor::registerExec()
+ 35.88% mongo::PlanExecutor::deregisterExec()
+ 5.40% mongo::runQuery(mongo::OperationContext*, mongo::QueryMessage&, mongo::Namesp
+ 5.26% mongo::AutoGetCollection::AutoGetCollection(mongo::OperationContext*, mongo::
+ 4.98% mongo::AutoGetCollectionForRead::~AutoGetCollectionForRead()
+ 2.61% mongo::getExecutorFind(mongo::OperationContext*, mongo::Collection*, mongo::N
+ 1.92% mongo::WiredTigerCursor::WiredTigerCursor(std::string const&, unsigned long,
+ 1.72% mongo::WiredTigerRecoveryUnit::~WiredTigerRecoveryUnit()
+ 1.09% mongo::fillOutPlannerParams(mongo::OperationContext*, mongo::Collection*, mon
+ 30.15% futex_wait_setup
+ 18.51% try_to_wake_up
vs:
benchRun.find 32 true 30
119839 ops/s
- 43.11% mongod [kernel.kallsyms] [k] _raw_spin_lock
- _raw_spin_lock
- 51.98% futex_wake
do_futex
sys_futex
system_call_fastpath
- __lll_unlock_wake
+ 23.98% mongo::ClientCursor::ClientCursor(mongo::CursorManager*, mongo::PlanExecutor
+ 20.36% mongo::ClientCursorPin::deleteUnderlying()
+ 19.67% mongo::PlanExecutor::registerExec()
+ 18.84% mongo::PlanExecutor::deregisterExec()
+ 16.71% mongo::ClientCursorPin::ClientCursorPin(mongo::CursorManager*, long long)
- 45.15% futex_wait_setup
futex_wait
do_futex
sys_futex
system_call_fastpath
- __lll_lock_wait
+ 22.64% mongo::ClientCursor::ClientCursor(mongo::CursorManager*, mongo::PlanExecutor
+ 22.05% mongo::PlanExecutor::deregisterExec()
+ 19.39% mongo::PlanExecutor::registerExec()
+ 18.94% mongo::ClientCursorPin::deleteUnderlying()
+ 16.70% mongo::ClientCursorPin::ClientCursorPin(mongo::CursorManager*, long long)
+ 2.56% try_to_wake_up
Contention appears to be CursorManager::_mutex.
- is related to
-
SERVER-20596 Performance regression in new mongos query path (about 17% worse than 3.0)
-
- Closed
-