MogDB丟失DW（double write）文件應急處理方法

原創范計杰 2022-09-05

1960

[[toc]]

適用范圍

當文件系統異常導致double write文件丟失或損壞，在沒有備份或者其它更合適的處理辦法時，可嘗試該文檔中的方法應急啟動數據庫。

問題概述

當開啟double write特性（默認開啟），丟失double write文件后數據庫無法啟動。
數據庫啟動失敗，日志如下，提示DW file文件不存在，錯誤堆棧時在做DW初始化。
2022-06-20 16:48:06.687 [unknown] [unknown] localhost 140123442810048 0 0 [BACKEND] LOG: start create thread!
2022-06-20 16:48:06.687 [unknown] [unknown] localhost 140123442810048 0 0 [BACKEND] LOG: create thread end!
2022-06-20 16:48:06.693 [unknown] [unknown] localhost 140122435614464 0 0 [BACKEND] LOG: [Alarm Module]alarm checker started.
2022-06-20 16:48:06.694 [unknown] [unknown] localhost 140122416674560 0 0 [BACKEND] LOG: reaper backend started.
2022-06-20 16:48:06.717 [unknown] [unknown] localhost 140122357491456 0 0 [REDO] LOG: [mpfl_ulink_file]: unlink global/max_page_flush_lsn sucessfully! ret:4294967295
2022-06-20 16:48:06.717 [unknown] [unknown] localhost 140122357491456 0 0 [BACKEND] LOG: StartupXLOG: biggest_lsn_in_page is set to FFFFFFFF/FFFFFFFF, enable_update_max_page_flush_lsn:0
2022-06-20 16:48:06.717 [unknown] [unknown] localhost 140122357491456 0 0 [BACKEND] LOG: database system timeline: 17
2022-06-20 16:48:06.717 [unknown] [unknown] localhost 140122357491456 0 0 [BACKEND] LOG: database system was shut down at 2022-06-20 16:47:32 CST
2022-06-20 16:48:06.720 [unknown] [unknown] localhost 140122357491456 0 0 [DBL_WRT] - [ ] PANIC: batch flush DW file does not exist <<<<<
2022-06-20 16:48:06.720 [unknown] [unknown] localhost 140122357491456 0 0 [DBL_WRT] BACKTRACELOG: tid[2843]'s backtrace:
/opt/og/bin/gaussdb(+0x9f16e2) [0x5625215106e2]
/opt/og/bin/gaussdb(_Z9errfinishiz+0x31c) [0x56252150274c]
/opt/og/bin/gaussdb(_Z25dw_file_check_and_rebuildv+0x128) [0x562521da4c68]
/opt/og/bin/gaussdb(_Z7dw_initb+0x7d) [0x562521da870d]
/opt/og/bin/gaussdb(_Z11StartupXLOGv+0x177b) [0x562521dddfbb]
/opt/og/bin/gaussdb(_Z18StartupProcessMainv+0x1ac) [0x5625219add4c]
/opt/og/bin/gaussdb(_Z26GaussDbAuxiliaryThreadMainIL15knl_thread_role26EEiP14knl_thread_arg+0xe0) [0x5625219a8f20]
/opt/og/bin/gaussdb(_Z17GaussDbThreadMainIL15knl_thread_role26EEiP14knl_thread_arg+0x245) [0x5625219a9185]
/opt/og/bin/gaussdb(+0xe6dc25) [0x56252198cc25]
/lib64/libpthread.so.0(+0x7e65) [0x7f70ffbebe65]
/lib64/libc.so.6(clone+0x6d) [0x7f70ff91488d]
Use addr2line to get pretty function name and line

問題原因

啟動時需要初始化double write，必要時使用DW RECOVER。由于DW文件丟失啟動失敗。

解決方案

1、重新初始化一個臨時cluster，復制新生成的dw文件到需要修復的cluster中
$ gs_initdb -D ./tmpdata --nodename test
$ ls tmpdata/global/pg_dw*
tmpdata/global/pg_dw tmpdata/global/pg_dw_single
$ cp ./tmpdata/global/pg_dw* data/global/
2、修改參數文件，設置enable_double_write = off
這時啟動數據庫可以正常啟動。
注意：
當需要dw處理fracture page時可能會丟數據。