MongoDB 查询在 Global.timeAcquiringMicros.r 上花费大量时间
MongoDB queries spending lots of time on Global.timeAcquiringMicros.r
我正在使用 MongoDB 版本 3.0.14(带有 wiredtiger)作为 4 成员回复集。
我在生产中遇到了一个奇怪的问题,突然大多数查询开始在 Global.timeAcquiringMicros.r
上阻塞(在辅助 Mongod 服务器上,java-客户端正在使用 SecondaryPreferred Read Preference 发送正常的读取查询)
2020-06-09T12:30:23.959+0000 I COMMAND [conn210676] query <db>.<collection> query: { $query: { _id: { $gte: "<value>", $lt: "<value>" } }, $orderby: { _id: -1 }, $maxTimeMS: 16000 } planSummary: IXSCAN { _id: 1 } ntoreturn:0 ntoskip:0 nscanned:11 nscannedObjects:11 keyUpdates:0 writeConflicts:0 numYields:6 nreturned:11 reslen:270641 locks:{ Global: { acquireCount: { r: 14 }, acquireWaitCount: { r: 1 }, timeAcquiringMicros: { r: 4580895 } }, Database: { acquireCount: { r: 7 } }, Collection: { acquireCount: { r: 7 } } } 1706ms
2020-06-09T12:30:25.887+0000 I COMMAND [conn210607] query <db>.<collection> query: { $query: { _id: { $gte: "<value1>", $lt: "<value2>" } }, $orderby: { _id: -1 }, $maxTimeMS: 16000 } planSummary: IXSCAN { _id: 1 } cursorid:76676946055 ntoreturn:0 ntoskip:0 nscanned:40 nscannedObjects:40 keyUpdates:0 writeConflicts:0 numYields:12 nreturned:40 reslen:1062302 locks:{ Global: { acquireCount: { r: 26 }, acquireWaitCount: { r: 1 }, timeAcquiringMicros: { r: 21622755 } }, Database: { acquireCount: { r: 13 } }, Collection: { acquireCount: { r: 13 } } } 3639ms
这些查询正在获取全局数据库上的意向共享锁 (r),它必须等待 4580895/21622755 微秒(根据 timeAcquiringMicros)。我有以下疑问:
- 它必须花费 4580895/21622755 微秒,这是怎么回事?我可以保证除了基于
_id
索引 的正常 delete/update/insert/query 之外没有任何活动发生
- 由于
Global.timeAcquiringMicros.r
很高,我假设有任何操作对全局数据库进行'R','W'锁定,那么,捕获此类查询的方法是什么?我已经尝试 db.currentOp()
但找不到任何东西。
- Global.acquireCount=14,为什么查询要取全局意向共享锁(r)14次?
解释第二个查询的结果(Global.timeAcquiringMicros.r = 21622755
):
{
"queryPlanner" : {
"plannerVersion" : 1,
"namespace" : "<db>.<coll>",
"indexFilterSet" : false,
"parsedQuery" : {
"$and" : [
{
"_id" : {
"$lt" : "<stop>"
}
},
{
"_id" : {
"$gte" : "<start>"
}
}
]
},
"winningPlan" : {
"stage" : "FETCH",
"inputStage" : {
"stage" : "IXSCAN",
"keyPattern" : {
"_id" : 1
},
"indexName" : "_id_",
"isMultiKey" : false,
"direction" : "backward",
"indexBounds" : {
"_id" : [
"(\"<stop>\", \"<start>\"]"
]
}
}
},
"rejectedPlans" : [ ]
},
"executionStats" : {
"executionSuccess" : true,
"nReturned" : 69,
"executionTimeMillis" : 57,
"totalKeysExamined" : 69,
"totalDocsExamined" : 69,
"executionStages" : {
"stage" : "FETCH",
"nReturned" : 69,
"executionTimeMillisEstimate" : 50,
"works" : 70,
"advanced" : 69,
"needTime" : 0,
"needFetch" : 0,
"saveState" : 2,
"restoreState" : 2,
"isEOF" : 1,
"invalidates" : 0,
"docsExamined" : 69,
"alreadyHasObj" : 0,
"inputStage" : {
"stage" : "IXSCAN",
"nReturned" : 69,
"executionTimeMillisEstimate" : 0,
"works" : 70,
"advanced" : 69,
"needTime" : 0,
"needFetch" : 0,
"saveState" : 2,
"restoreState" : 2,
"isEOF" : 1,
"invalidates" : 0,
"keyPattern" : {
"_id" : 1
},
"indexName" : "_id_",
"isMultiKey" : false,
"direction" : "backward",
"indexBounds" : {
"_id" : [
"(\"<stop>\", \"<start>\"]"
]
},
"keysExamined" : 69,
"dupsTested" : 0,
"dupsDropped" : 0,
"seenInvalidated" : 0,
"matchTested" : 0
}
}
},
"serverInfo" : {
"host" : "<host>",
"port" : 27017,
"version" : "3.0.14",
"gitVersion" : "08352afcca24bfc145240a0fac9d28b978ab77f3"
},
"ok" : 1
}
当这种行为发生时,我能够捕获 currentOp() 之一:
{
"desc" : "conn225729",
"threadId" : "0x1321c1400",
"connectionId" : 225729,
"opid" : 189970948,
"active" : false,
"op" : "getmore",
"ns" : "<db>.<coll>",
"query" : {
},
"client" : "<client-ip>:55596",
"numYields" : 0,
"locks" : {
"Global" : "r"
},
"waitingForLock" : true,
"lockStats" : {
"Global" : {
"acquireCount" : {
"r" : NumberLong(1)
},
"acquireWaitCount" : {
"r" : NumberLong(1)
},
"timeAcquiringMicros" : {
"r" : NumberLong(7500907)
}
}
}
}
- 这里的问题也是一样的,正常查询等待
Global.timeAcquiringMicros.r
("waitingForLock" : true)
请帮助并参考一些 MongoDB 解释问题的文档。另外,如果需要任何其他日志,请告诉我。
A MongoDB 辅助节点从主节点批量检索操作日志事件。当它应用一批 oplog 事件时,它需要一个全局独占写锁 (W)。
读意图锁(r) 与W
锁互斥。
这意味着写入和读取必须在辅助节点上交错,因此大量写入会阻塞读取,大量读取会延迟复制。
Non-blocking secondary reads 是几年前 MongoDB 4.0 的主要功能。
如果您可以升级,则不会再发生特定的锁争用。
我正在使用 MongoDB 版本 3.0.14(带有 wiredtiger)作为 4 成员回复集。
我在生产中遇到了一个奇怪的问题,突然大多数查询开始在 Global.timeAcquiringMicros.r
上阻塞(在辅助 Mongod 服务器上,java-客户端正在使用 SecondaryPreferred Read Preference 发送正常的读取查询)
2020-06-09T12:30:23.959+0000 I COMMAND [conn210676] query <db>.<collection> query: { $query: { _id: { $gte: "<value>", $lt: "<value>" } }, $orderby: { _id: -1 }, $maxTimeMS: 16000 } planSummary: IXSCAN { _id: 1 } ntoreturn:0 ntoskip:0 nscanned:11 nscannedObjects:11 keyUpdates:0 writeConflicts:0 numYields:6 nreturned:11 reslen:270641 locks:{ Global: { acquireCount: { r: 14 }, acquireWaitCount: { r: 1 }, timeAcquiringMicros: { r: 4580895 } }, Database: { acquireCount: { r: 7 } }, Collection: { acquireCount: { r: 7 } } } 1706ms
2020-06-09T12:30:25.887+0000 I COMMAND [conn210607] query <db>.<collection> query: { $query: { _id: { $gte: "<value1>", $lt: "<value2>" } }, $orderby: { _id: -1 }, $maxTimeMS: 16000 } planSummary: IXSCAN { _id: 1 } cursorid:76676946055 ntoreturn:0 ntoskip:0 nscanned:40 nscannedObjects:40 keyUpdates:0 writeConflicts:0 numYields:12 nreturned:40 reslen:1062302 locks:{ Global: { acquireCount: { r: 26 }, acquireWaitCount: { r: 1 }, timeAcquiringMicros: { r: 21622755 } }, Database: { acquireCount: { r: 13 } }, Collection: { acquireCount: { r: 13 } } } 3639ms
这些查询正在获取全局数据库上的意向共享锁 (r),它必须等待 4580895/21622755 微秒(根据 timeAcquiringMicros)。我有以下疑问:
- 它必须花费 4580895/21622755 微秒,这是怎么回事?我可以保证除了基于
_id
索引 的正常 delete/update/insert/query 之外没有任何活动发生
- 由于
Global.timeAcquiringMicros.r
很高,我假设有任何操作对全局数据库进行'R','W'锁定,那么,捕获此类查询的方法是什么?我已经尝试db.currentOp()
但找不到任何东西。 - Global.acquireCount=14,为什么查询要取全局意向共享锁(r)14次?
解释第二个查询的结果(Global.timeAcquiringMicros.r = 21622755
):
{
"queryPlanner" : {
"plannerVersion" : 1,
"namespace" : "<db>.<coll>",
"indexFilterSet" : false,
"parsedQuery" : {
"$and" : [
{
"_id" : {
"$lt" : "<stop>"
}
},
{
"_id" : {
"$gte" : "<start>"
}
}
]
},
"winningPlan" : {
"stage" : "FETCH",
"inputStage" : {
"stage" : "IXSCAN",
"keyPattern" : {
"_id" : 1
},
"indexName" : "_id_",
"isMultiKey" : false,
"direction" : "backward",
"indexBounds" : {
"_id" : [
"(\"<stop>\", \"<start>\"]"
]
}
}
},
"rejectedPlans" : [ ]
},
"executionStats" : {
"executionSuccess" : true,
"nReturned" : 69,
"executionTimeMillis" : 57,
"totalKeysExamined" : 69,
"totalDocsExamined" : 69,
"executionStages" : {
"stage" : "FETCH",
"nReturned" : 69,
"executionTimeMillisEstimate" : 50,
"works" : 70,
"advanced" : 69,
"needTime" : 0,
"needFetch" : 0,
"saveState" : 2,
"restoreState" : 2,
"isEOF" : 1,
"invalidates" : 0,
"docsExamined" : 69,
"alreadyHasObj" : 0,
"inputStage" : {
"stage" : "IXSCAN",
"nReturned" : 69,
"executionTimeMillisEstimate" : 0,
"works" : 70,
"advanced" : 69,
"needTime" : 0,
"needFetch" : 0,
"saveState" : 2,
"restoreState" : 2,
"isEOF" : 1,
"invalidates" : 0,
"keyPattern" : {
"_id" : 1
},
"indexName" : "_id_",
"isMultiKey" : false,
"direction" : "backward",
"indexBounds" : {
"_id" : [
"(\"<stop>\", \"<start>\"]"
]
},
"keysExamined" : 69,
"dupsTested" : 0,
"dupsDropped" : 0,
"seenInvalidated" : 0,
"matchTested" : 0
}
}
},
"serverInfo" : {
"host" : "<host>",
"port" : 27017,
"version" : "3.0.14",
"gitVersion" : "08352afcca24bfc145240a0fac9d28b978ab77f3"
},
"ok" : 1
}
当这种行为发生时,我能够捕获 currentOp() 之一:
{
"desc" : "conn225729",
"threadId" : "0x1321c1400",
"connectionId" : 225729,
"opid" : 189970948,
"active" : false,
"op" : "getmore",
"ns" : "<db>.<coll>",
"query" : {
},
"client" : "<client-ip>:55596",
"numYields" : 0,
"locks" : {
"Global" : "r"
},
"waitingForLock" : true,
"lockStats" : {
"Global" : {
"acquireCount" : {
"r" : NumberLong(1)
},
"acquireWaitCount" : {
"r" : NumberLong(1)
},
"timeAcquiringMicros" : {
"r" : NumberLong(7500907)
}
}
}
}
- 这里的问题也是一样的,正常查询等待
Global.timeAcquiringMicros.r
("waitingForLock" : true)
请帮助并参考一些 MongoDB 解释问题的文档。另外,如果需要任何其他日志,请告诉我。
A MongoDB 辅助节点从主节点批量检索操作日志事件。当它应用一批 oplog 事件时,它需要一个全局独占写锁 (W)。
读意图锁(r) 与W
锁互斥。
这意味着写入和读取必须在辅助节点上交错,因此大量写入会阻塞读取,大量读取会延迟复制。
Non-blocking secondary reads 是几年前 MongoDB 4.0 的主要功能。
如果您可以升级,则不会再发生特定的锁争用。