go sql driver + context canceled 似乎是 "leak"goroutines？

Question

背景

当我对我正在构建的应用程序运行负载测试时，一切都运行顺利，直到机器几乎耗尽内存并且应用程序遇到问题。突然所有上下文都被取消了(很可能被 hystrix 库取消了)，因为处理请求的时间太长了。到目前为止，没有什么真正令人惊讶的。

问题

我确实觉得奇怪的是，当上下文被取消时，它似乎开始泄漏 go 例程。

在健康状况下，大约有 20 个 goroutine 一直在运行。但是在这种情况发生之后，它会保留 20 多个 goroutines:准确地说是 98 个。随着时间的推移，这个数字不会下降。我使用带有读/写超时的 http.Server，带有上下文的 go-sql-driver/mysql 和 hystrix。下面是 goroutines 的转储，这些 goroutines 的运行时间比所有超时时间都长。几乎所有这些都来自 go-sql-driver 或由 go-sql-driver 调用。 MySQL 进程列表中没有剩余查询。想知道其他人是否遇到过这种情况并知道如何解决。

此后应用程序仍然运行流畅，可以处理超过 2000r/s，成功率 100%，但在测试完成后，之前挂起的 goroutines 仍然存在。

这是 goroutine 的转储(不是全部，但大体相同)。 runtime_pollWait 来自 go-mysql-driver:

goroutine 1 [IO wait]:
internal/poll.runtime_pollWait(0x7fd0fbc007b8, 0x72, 0x0)

goroutine 18 [sleep, 43 minutes]:
runtime.goparkunlock(...)

goroutine 19 [select, 2 minutes]:
database/sql.(*DB).connectionOpener(0xc000188240, 0xb835a0, 0xc00017b2c0)

goroutine 20 [select]:
database/sql.(*DB).connectionResetter(0xc000188240, 0xb835a0, 0xc00017b2c0)

goroutine 34 [select, 2 minutes]:
database/sql.(*DB).connectionOpener(0xc0001883c0, 0xb835a0, 0xc00016a140)

goroutine 35 [select]:
database/sql.(*DB).connectionResetter(0xc0001883c0, 0xb835a0, 0xc00016a140)

goroutine 8 [chan receive]:
github.com/go-redis/redis/internal/pool.(*ConnPool).reaper(0xc0000ba370, 0xdf8475800)

goroutine 11 [chan receive]:
github.com/go-redis/redis/internal/pool.(*ConnPool).reaper(0xc0000ba420, 0xdf8475800)

goroutine 14 [chan receive]:
github.com/go-redis/redis/internal/pool.(*ConnPool).reaper(0xc0000ba4d0, 0xdf8475800)

goroutine 98 [sleep, 43 minutes]:
runtime.goparkunlock(...)

goroutine 87 [chan receive]:
github.com/afex/hystrix-go/hystrix.(*metricExchange).Monitor(0xc002a1b5c0)

goroutine 99 [syscall]:
syscall.Syscall6(0xe8, 0xd, 0xc0002b9a58, 0x64, 0x64, 0x0, 0x0, 0x42bfcf, 0x8, 0xacdd80)

goroutine 100 [sleep]:
runtime.goparkunlock(...)

goroutine 88 [chan receive]:
github.com/afex/hystrix-go/hystrix.(*poolMetrics).Monitor(0xc003ac08a0)

goroutine 91 [chan receive]:
github.com/afex/hystrix-go/hystrix.(*metricExchange).Monitor(0xc002a1b680)

goroutine 92 [chan receive]:
github.com/afex/hystrix-go/hystrix.(*poolMetrics).Monitor(0xc003ac0f00)

goroutine 25921733 [IO wait, 3 minutes]:
internal/poll.runtime_pollWait(0x7fd0fa06f298, 0x72, 0xffffffffffffffff)

goroutine 25921439 [IO wait, 3 minutes]:
internal/poll.runtime_pollWait(0x7fd0f983fe18, 0x72, 0xffffffffffffffff)

goroutine 25927120 [select, 3 minutes]:
github.com/go-sql-driver/mysql.(*mysqlConn).startWatcher.func1(0xc003aa7620, 0xc01c2ef980, 0xc01a841bc0)

goroutine 25928175 [select, 3 minutes]:
github.com/go-sql-driver/mysql.(*mysqlConn).startWatcher.func1(0xc012abbf80, 0xc01c2ee600, 0xc01144f8c0)

goroutine 25927727 [select, 3 minutes]:
github.com/go-sql-driver/mysql.(*mysqlConn).startWatcher.func1(0xc014ae4300, 0xc0166a3c80, 0xc0030e8a80)

goroutine 24567262 [select]:
database/sql.(*DB).connectionCleaner(0xc000188240, 0x2540be400)

goroutine 25927990 [select, 3 minutes]:
github.com/go-sql-driver/mysql.(*mysqlConn).startWatcher.func1(0xc012ababa0, 0xc0166a3ec0, 0xc01144f380)

goroutine 25921926 [IO wait, 3 minutes]:
internal/poll.runtime_pollWait(0x7fd0fa0702d8, 0x72, 0xffffffffffffffff)

goroutine 25921958 [IO wait, 3 minutes]:
internal/poll.runtime_pollWait(0x7fd0fa06eb48, 0x72, 0xffffffffffffffff)

goroutine 24567381 [select]:
database/sql.(*DB).connectionCleaner(0xc0001883c0, 0x2540be400)

goroutine 25921798 [IO wait, 3 minutes]:
internal/poll.runtime_pollWait(0x7fd0f9882038, 0x72, 0xffffffffffffffff)

goroutine 25921751 [IO wait, 3 minutes]:
internal/poll.runtime_pollWait(0x7fd0fa6b6820, 0x72, 0xffffffffffffffff)

goroutine 25921757 [IO wait, 3 minutes]:
internal/poll.runtime_pollWait(0x7fd0fa6b6410, 0x72, 0xffffffffffffffff)

goroutine 25927113 [select, 3 minutes]:
github.com/go-sql-driver/mysql.(*mysqlConn).startWatcher.func1(0xc003aa7320, 0xc01c2efbc0, 0xc01a8419e0)

goroutine 25921966 [IO wait, 3 minutes]:
internal/poll.runtime_pollWait(0x7fd0fa06e8d8, 0x72, 0xffffffffffffffff)

goroutine 25921964 [IO wait, 3 minutes]:
internal/poll.runtime_pollWait(0x7fd0fa06e9a8, 0x72, 0xffffffffffffffff)

goroutine 25921323 [IO wait, 3 minutes]:
internal/poll.runtime_pollWait(0x7fd0fa06f0f8, 0x72, 0xffffffffffffffff)

goroutine 25928466 [select, 3 minutes]:
github.com/go-sql-driver/mysql.(*mysqlConn).startWatcher.func1(0xc00e552240, 0xc01c2ef740, 0xc01144f920)

goroutine 25928011 [select, 3 minutes]:
github.com/go-sql-driver/mysql.(*mysqlConn).startWatcher.func1(0xc0156599e0, 0xc01c2ee3c0, 0xc01baa8fc0)

goroutine 25919812 [IO wait, 3 minutes]:
internal/poll.runtime_pollWait(0x7fd0fa06ea78, 0x72, 0xffffffffffffffff)

goroutine 25928014 [select, 3 minutes]:
github.com/go-sql-driver/mysql.(*mysqlConn).startWatcher.func1(0xc015659b00, 0xc01177e900, 0xc01baa90e0)

goroutine 25922001 [IO wait, 3 minutes]:
internal/poll.runtime_pollWait(0x7fd0fa070548, 0x72, 0xffffffffffffffff)

goroutine 25925675 [IO wait, 3 minutes]:
internal/poll.runtime_pollWait(0x7fd0fa6b8150, 0x72, 0xffffffffffffffff)

goroutine 25928529 [select, 3 minutes]:
github.com/go-sql-driver/mysql.(*mysqlConn).startWatcher.func1(0xc003aa7680, 0xc01c2ef8c0, 0xc01a841c20)

goroutine 25928467 [select, 3 minutes]:
github.com/go-sql-driver/mysql.(*mysqlConn).startWatcher.func1(0xc00e552360, 0xc01c2ef680, 0xc01144f980)

goroutine 25928468 [select, 3 minutes]:
github.com/go-sql-driver/mysql.(*mysqlConn).startWatcher.func1(0xc00e552480, 0xc01c2ef5c0, 0xc01144f9e0)

goroutine 25919817 [IO wait, 3 minutes]:
internal/poll.runtime_pollWait(0x7fd0fa06e808, 0x72, 0xffffffffffffffff)

goroutine 25925511 [IO wait, 3 minutes]:
internal/poll.runtime_pollWait(0x7fd0f983e4e8, 0x72, 0xffffffffffffffff)

goroutine 25920658 [IO wait, 3 minutes]:
internal/poll.runtime_pollWait(0x7fd0fa070208, 0x72, 0xffffffffffffffff)

goroutine 25920661 [IO wait, 3 minutes]:
internal/poll.runtime_pollWait(0x7fd0fa070138, 0x72, 0xffffffffffffffff)

goroutine 25920667 [IO wait, 3 minutes]:
internal/poll.runtime_pollWait(0x7fd0fa06e738, 0x72, 0xffffffffffffffff)

goroutine 25920860 [IO wait, 3 minutes]:
internal/poll.runtime_pollWait(0x7fd0fa06f368, 0x72, 0xffffffffffffffff)

goroutine 25920790 [IO wait, 3 minutes]:
internal/poll.runtime_pollWait(0x7fd0fa070478, 0x72, 0xffffffffffffffff)

goroutine 25920793 [IO wait, 3 minutes]:
internal/poll.runtime_pollWait(0x7fd0fa06fec8, 0x72, 0xffffffffffffffff)

goroutine 25928006 [select, 3 minutes]:
github.com/go-sql-driver/mysql.(*mysqlConn).startWatcher.func1(0xc015659800, 0xc01c1603c0, 0xc01baa8de0)

goroutine 25921085 [IO wait, 3 minutes]:
internal/poll.runtime_pollWait(0x7fd0fa6b7fb0, 0x72, 0xffffffffffffffff)

.....

goroutine 25925779 [IO wait, 3 minutes]:
internal/poll.runtime_pollWait(0x7fd0fa6b7ba0, 0x72, 0xffffffffffffffff)

goroutine 25926131 [IO wait, 3 minutes]:
internal/poll.runtime_pollWait(0x7fd0fa6b69c0, 0x72, 0xffffffffffffffff)

goroutine 25927988 [select, 3 minutes]:
github.com/go-sql-driver/mysql.(*mysqlConn).startWatcher.func1(0xc012abaae0, 0xc01c2ee000, 0xc01144f2c0)

goroutine 25928470 [select, 3 minutes]:
github.com/go-sql-driver/mysql.(*mysqlConn).startWatcher.func1(0xc00e5526c0, 0xc01c2ef440, 0xc01144faa0)

goroutine 25928471 [select, 3 minutes]:
github.com/go-sql-driver/mysql.(*mysqlConn).startWatcher.func1(0xc00e5527e0, 0xc01c2ef380, 0xc01144fb00)

goroutine 25928472 [select, 3 minutes]:
github.com/go-sql-driver/mysql.(*mysqlConn).startWatcher.func1(0xc00e552900, 0xc01c2ef2c0, 0xc01144fb60)

goroutine 25928473 [select, 3 minutes]:
github.com/go-sql-driver/mysql.(*mysqlConn).startWatcher.func1(0xc00e552a20, 0xc01c2ef200, 0xc01144fbc0)

goroutine 26252345 [IO wait]:
internal/poll.runtime_pollWait(0x7fd0e7054fe8, 0x72, 0xffffffffffffffff)

Answer 1

回答我自己的问题；问题很可能是系统过载。 我将并发连接数限制在 25% 之内，并且使用的 MySQL 连接数少了很多。这似乎解决了问题。我现在的吞吐量较少，但系统更稳定。

它仍然没有回答为什么即使系统再次平静下来，netpoller 仍然阻塞。他们只是永远不会恢复，我认为这不好。但可能有一个很好的理由。我在 Stackoverflow 上通过网络连接多次看到这种症状，例如 http.Client。