v3.1.1 coredump 在slash::PosixMmapFile::Append

[hotrhino]

#0 0x00007f8a5042bb0e in __memcpy_ssse3_back () from /lib64/libc.so.6
#1 0x00000000006828e5 in slash::PosixMmapFile::Append (this=0x7f89f801f4d0, data=...) at src/env.cc:512
#2 0x000000000068a157 in slash::BaseConf::WriteBack (this=this@entry=0x19706d0) at src/base_conf.cc:270
#3 0x0000000000606359 in PikaConf::ConfigRewrite (this=0x19706d0) at /data/pika_src/pika/src/pika_conf.cc:410
#4 0x00000000005d21b4 in ConfigRewrite (ret="", this=0x7f89f8001ab0) at /data/pika_src/pika/src/pika_admin.cc:1593
#5 ConfigCmd::Do (this=0x7f89f8001ab0, partition=...) at /data/pika_src/pika/src/pika_admin.cc:1005
#6 0x00000000005a2d98 in Cmd::ProcessDoNotSpecifyPartitionCmd (this=) at /data/pika_src/pika/src/pika_command.cc:547
#7 0x000000000061cd90 in PikaClientConn::DoCmd (this=0x7f8a20000908, argv=std::vector of length 2, capacity 2 = {...}, opt="config") at /data/pika_src/pika/src/pika_client_conn.cc:192
#8 0x000000000061dca1 in PikaClientConn::DealMessage (this=0x7f8a20000908, argv=std::vector of length 2, capacity 2 = {...}, response=0x7f8a20000980) at /data/pika_src/pika/src/pika_client_conn.cc:253
#9 0x000000000061a6f9 in PikaClientConn::BatchExecRedisCmd (this=0x7f8a20000908, argvs=std::vector of length 1, capacity 1 = {...}, response=0x7f8a20000980) at /data/pika_src/pika/src/pika_client_conn.cc:234
#10 0x000000000061a774 in PikaClientConn::DoBackgroundTask (arg=0x7f8a20009e20) at /data/pika_src/pika/src/pika_client_conn.cc:263
#11 0x000000000067fff7 in pink::ThreadPool::runInThread (this=0x197a4b0) at src/thread_pool.cc:183
#12 0x00000000006801b9 in pink::ThreadPool::Worker::WorkerMain (arg=) at src/thread_pool.cc:14
#13 0x00007f8a51710e25 in start_thread () from /lib64/libpthread.so.0
#14 0x00007f8a503d5bad in clone () from /lib64/libc.so.6

centos7.5 原先使用预编译的pika-linux-x86_64-v3.1.1.tar 发生coredump后 因为是提示内存对齐方面的错误，git拉源码切到v3.1.1编译 进行运行一会后照样coredump在同一个位置。

使用标准配置， 我有个三个redis库想合并到同一个pika里面，使用工具把三个库导入后，对每个库都使用redis-copy命令 及redis_instA<==>pika_instA redis_instB<==>pika_instA redis_instC<==>pika_instA, redis-copy三对一。 运行一会后coredump

[hotrhino]

大概找到问题了 没有时间去深入源码 库维护人员看下吧。
基本都是在int PikaConf::ConfigRewrite()的时候coredump，里面最后一行WriteBack没有在锁范围内，里面有创建mmap文件的操作，WriteBack如果并行执行必然会导致问题。我把WriteBack纳入RWLock l(&rwlock_, true)内后不再coredump了

[Axlgrep]

好的， 我们跟一下

[Axlgrep]

#686 fixed

[hotrhino]

好的 多谢。