Pciessd异常Readonly致Mysql反复crash如何处理？
去年10月份发生一起fio卡变为readonly(和双十一无关)，发生一起fio卡变为readonly，造成mysql crash的故障，整理如下。
【机器配置】
- System | Dell Inc.; PowerEdge R710;
- Processors | physical = 2, cores = 12, virtual = 24, hyperthreading = yes
- # Memory #####################################################
- Total | 94.40G
- Free | 555.50M
- Swappiness | vm.swappiness = 0
- # Disk #####################################################
- 2 ioMemory devices in this system
- Fusion-io driver version: 3.1.5 build 126
- Fusion-io ioDrive 640GB *2 –> mdadm –>/dev/md0
- ibdata,ib_logfile,bin_log,relay_log on SAS 600GB raid1
【问题表现】
13:28，监控发出***个db_ping告警
mysql的alert log如下：
- /u01/mysql/libexec/mysqld: Can’t create/write to file ‘/u01/mysql/tmp/ibU5kXB4′ ( Errcode: 30)
- 121104 13:28:10 InnoDB: Error: unable to create temporary file; errno: 30
- 121104 13:28:10 [ERROR] Plugin ‘InnoDB’ init function returned error.
- 121104 13:28:10 [ERROR] Plugin ‘InnoDB’ registration as a STORAGE ENGINE failed.
- 121104 13:28:10 [ERROR] Aborting
- InnoDB: Error: tried to read 16384 bytes at offset 0 41517056.
- InnoDB: Was only able to read -1.
- 121104 13:14:59 InnoDB: Operating system error number 5 in a file operation.
- InnoDB: Error number 5 means ‘Input/output error’.
- InnoDB: Some operating system error numbers are described at
- InnoDB: http://dev.mysql.com/doc/refman/5.1/en/operating-system-error-codes.html
- InnoDB: File operation call: ‘read’.
- InnoDB: Cannot continue operation.
- mysqld: my_new.cc:51: int __cxa_pure_virtual(): Assertion `! “Aborted: pure virtual method called.”‘ failed.
- 121104 13:14:59 – mysqld got signal 6 ;
由上判断IO设备有问题，此时touch /u01/mysql/tmp/ibd： touch: cannot touch `/u01/mysql/tmp/ibd’: Read-only file system
[SITESERVER_PAGE]由于是核心集群，有数据强一致需求，通过DBA手工强制主备切换，故障排除。
【问题原因】
- fusionIO卡出现readonly
- /var/log/message
- Nov 4 13:14:59 my160130.cm6 kernel: : fioerr Fusion-io ioDrive 640GB 0000:07:00.0: Single Bit Event Upset Error Dete4ted – interrupt: val[0]: 000ff16
- fio-status -a
- fct1 Failed: DEVICE IS OFFLINE. ALL READS AND WRITES WILL FAIL!
- ioDrive 640GB MLC, Product Number:2TTK9, SN:436946
- !! —> There are active errors or warnings on this device! Read below for details.
- ioDrive 640GB MLC, PN:00214401201
- Located in slot 0 Center of Pseudo Low-Profile ioDIMM Adapter SN:436946
- WARNING: READ-ONLY MODE. ALL WRITES WILL FAIL!
- ACTIVE ERRORS:
- The ioMemory has encountered an internal error and has been
- temporarily disabled. All reads and writes will fail.
- The ioMemory is not allowing write operations.
【问题分析】
[SITESERVER_PAGE]•SEUs are transient soft errors, and are non-destructive. A reset or rewriting of the device results in normal device behavior thereafter
fio的控制模块是跑在fpga上的，元数据存储在DRAM和SSD上，断电可恢复。2.x的驱动发生该错误后，会rewriting进行修复。3.x的驱动提高了安全性，发生该错误后，会直接reset，卡read_only等待power recycle
•SEU class errors are caused by cosmic ray particles making it’s way into the NAND controller or by a failing NAND controller
FPGA本身的介质损坏或者宇宙射线，都是该错误的诱因。五月份message中有类似Write Path报错，2.x驱动自动rewrite修复了，3.x的驱动安全级别更高，reset后置为readonly
•Write Path Parity Error
这个错误是SEU错误的前驱，绝大多数可修复。同集群中，有3台发生过并自动修复。
•FPGA的成本相比开芯片低廉很多，编程迭代迅速，但健壮性不开芯片
【数据丢失】
因undo,redo,binlog都在u02的SAS盘上日志完整，备库基本没有延迟，故没有数据丢失；
但由于SEU可能导致当时写入的block错误，造成data不一致，保险起见还是重做备库，利用binlog同步所有数据。
•SEU class of error my result in data on the device being corrupted.The database should be verified or restored from backup
【改进措施】
FPGA老化后，有一定几率发生Single Event Upset错误，核心库要及时替换；
FPGA对宇宙射线敏感，需要控制机房环境，并分散机柜上架；
改进更敏感的message,dmesg告警。