NetWorker:网络配置的最佳实践
摘要: 本文旨在为 NetWorker 主机的理想和标准网络可调参数提供一个简单的基线。
症状
- 与网络或主机连接相关的错误,包括但不限于:
- 看似已完成实际数据传输的备份失败
- 资源普遍枯竭或通信崩溃
GSS warning Session information (number hex:hex) registered by user for nsrexecd has expired because a NetWorker daemon had not requested it after 120 minutesGSS error Session information (number hex:hex) was requested by nsrmmd but the session has expiredRPC severe Unable to query NSR database for list of configured devices: RPC receive operation failed; peer = ip_addr:port, errno = Connection timed outRPC severe Unable to query NSR database for list of configured devices: RPC send operation failed; peer = ip_addr:port, errno = Broken pipeNSR notice Chunking ssid ssid failed, because saveset was abortedddp_open_file_ext() failed for File: //mtree/vol_dir/nn/nn/long_ssid, Err: 5004-nfs lookup failed (nfs: No such file or directory) ).NSR critical Connectivity check request is failed for: SN_CONN_REPORT_DD type data_domain deviceRPC error RPC client handle: No route to host.RPC error RPC client handle: Connection refused.RPC error Unable to create the connection with 'portmapper' to host 'hostname' with address 'ip_addr' at port number 7938.RPC critical Aborting client connection from ip_addr: Connection timed out.RPC critical Check whether the firewall is blocking the client ports on the host 'hostname'.RPC critical Check whether the client services are running on the host 'hostname'.
原因
NetWorker 是一种应用程序,可在常规作期间在本地和远程主机上创建许多套接字。虽然服务器和存储节点通常会创建更多内容,但客户端配置也可能影响作业成功。
保留者:NetWorker 调用方进程会创建套接字以连接到侦听器守护程序,但空闲连接可能会被回收未使用资源的网络设备中断。通常,这要求默认情况下为 NetWorker 服务器和节点以及遇到问题的客户端启用 keepalive。对于某些(但不是全部)二进制文件,NetWorker 有自己的内部 keepalive 处理。默认情况下,作系统还具有应启用的 keepalive。
端口可用性:NetWorker 套接字需要临时端口进行通信,但作系统默认限制此范围 — 将其完全扩展以避免限制或中断通信。使用 nsrauth 已启用(默认),每个插槽至少需要三个端口;失败会快速重试,使端口保持TIME_WAIT状态,直至成功建立连接。因此,应提高最大可用端口数,最好降低TIME_WAIT状态。
其他长时间运行的插槽也可以使用特定的内部软件变量进行强化,以实现更高的弹性或改进缓冲。
解决方案
以下是作系统和主机类的常见建议设置及其实现命令。总是,适用性各不相同;那些被认为普遍可取的不予注释,而那些适用性变化较大的则被注释,但可在需要时使用。这些设置作为一般建议真诚地提供,但在实施之前应由作系统管理员进行检视。这些被视为服务器和存储节点的最佳默认最佳实践。客户端适用性取决于配置和角色;冲突的服务器角色可能会覆盖建议的设置,因此在部署时应优先考虑特定于角色的要求。
Linux:所有适当的设置都应在 /nsr/nsrrc 文件,必须具有全局读取/执行权限 (755) 才能在服务启动时运行。默认标准条目未注释,注释非标准或间接选项。使用以下命令更改设置的可用性: # 相关行上的前缀。根据部署文件的位置,剪裁文件使其与 NetWorker 客户端、节点或服务器相关。进行更改后,将需要重新启动服务。
### GENERAL USAGE AND CONFIRMATION OF ENGAGEMENT # This file will only be read, parsed and executed if permissions are set to 755 EXACTLY. # Note that for Linux / Unix systems, it will only be engaged at service startup time, and will only # affect NetWorker runtime environment, overriding system settings, but leaving them in place for all # other operations. Changes to nsrrc require service restart to be engaged. To confirm parameters # are being engaged, check /nsr/nsrrc.log for datestamps and values. See end of script for details. ### LINUX - For all NetWorker hosts - Clients, Nodes and Server NSR_KEEPALIVE_WAIT=10 export NSR_KEEPALIVE_WAIT NSR_EXEC_MAX_AUTH_THREADS=50 export NSR_EXEC_MAX_AUTH_THREADS # NSR_SOCK_BUF_SIZE=65536 # (262144 for 10 Gb ETH NICs) # export NSR_SOCK_BUF_SIZE # NetWorker internal keepalive settings for some, but not all binaries - 4.5 minutes to ensure keepalives are passed before the increasingly common 5 minute router idle socket kill timer NW_TCP_KEEPIDLE_SECS=270 export NW_TCP_KEEPIDLE_SECS NW_TCP_KEEPINTVL_SECS=75 export NW_TCP_KEEPINTVL_SECS NW_TCP_KEEPCNT=20 export NW_TCP_KEEPCNT # OS-level keepalive values - also set to 4.5 minutes for the same reason sysctl -w "net.ipv4.tcp_keepalive_intvl=75" sysctl -w "net.ipv4.tcp_keepalive_probes=20" sysctl -w "net.ipv4.tcp_keepalive_time=270" # Set kernel limits to ensure core dump generation ulimit -Sn 262144 ulimit -Sc unlimited ### For NetWorker Storage Nodes and Server # Set kernel limits to provide maximum file descriptor availability ulimit -Hn 262144 ulimit -Hc unlimited # Globally disable IPv6, if it is not necessary for operation: # sysctl -w "net.ipv6.conf.all.disable_ipv6=1" # Disable dynamic TCP window scaling - requires compatible equipment in the data path, as well as ECN sysctl -w "net.ipv4.tcp_window_scaling=0" sysctl -w "net.ipv4.tcp_ecn=0" # Raise connection backlog (hash tables) to the maximum value allowed if desired # sysctl -w "net.ipv4.tcp_max_syn_backlog=8192" # sysctl -w "net.core.netdev_max_backlog=8192" # (For 10 Gb Eth use the value = 30000) # Raise memory size available for TCP buffers as needed # sysctl -w "net.core.rmem_default=262144" # sysctl -w "net.core.wmem_default=262144" # sysctl -w "net.core.rmem_max=16777216" # sysctl -w "net.core.wmem_max=16777216" # sysctl -w "net.ipv4.tcp_rmem=8192 524288 16777216" # sysctl -w "net.ipv4.tcp_wmem=8192 524288 16777216" # Increase shared memory pool if required - particularly for immediate mode on Storage Nodes # sysctl -w kernel.shmmax = 2147483648 # - e.g. 2 GB # sysctl -w kernel.shmall = 2147483648 # - e.g. 2 GB # Available TCP client ephemeral port range increase from default: sysctl -w "net.ipv4.ip_local_port_range=10000 64000" # Enable TCP Time Wait Reuse for very high load servers and nodes to increase socket reuse availability sysctl -w "net.ipv4.tcp_tw_recycle=0" sysctl -w "net.ipv4.tcp_tw_reuse=2" # Lower TIME_WAIT delay to close connections more quickly. This may not be necessary in concert with tw_reuse. # sysctl -w "net.ipv4.tcp_fin_timeout=30" # NFS I/O concurrency: sysctl -w "sunrpc.tcp_slot_table_entries=128" sysctl -w "sunrpc.udp_slot_table_entries=128" ### For NetWorker Server only # Settings to increase device resilience for cloud operations or other potentially high-latency devices # NSR_DEVOP_TIMEOUT=3600 # export NSR_DEVOP_TIMEOUT # NSR_DEVOP_POLLING_INTERVAL=600 # export NSR_DEVOP_POLLING_INTERVAL # NSR_DEVOP_INQUIRY_TIMEOUT=900 # export NSR_DEVOP_INQUIRY_TIMEOUT ### Media database tunables # NSR_TCP_READ_LONG_WAIT=Y # export NSR_TCP_READ_LONG_WAIT # NSR_MAX_MEDIADB_RETRY=10 # export NSR_MAX_MEDIADB_RETRY # MMDB_SQLITE_CONFIGURE_MEMORY=1 # export MMDB_SQLITE_CONFIGURE_MEMORY # MMDB_SQLITE_PAGECACHE_SIZE=65536 # export MMDB_SQLITE_PAGECACHE_SIZE # MMDB_SQLITE_PAGE_COUNT=65536 # export MMDB_SQLITE_PAGE_COUNT # MMDB_SQLITE_HEAP_SIZE=1073741824 # export MMDB_SQLITE_HEAP_SIZE # MDB_SQLITE_HEAP_MIN_ALLOC_SIZE=128 # export MDB_SQLITE_HEAP_MIN_ALLOC_SIZE ### NetWorker VMware Protection (NVP) Specific Tunables ## Increase the inventory (nsrvim) default timeout # GST_VBA_TIMEOUT=7200 # export GST_VBA_TIMEOUT # NSR_HYPERVISOR_QUERY_REQUEST_TIMEOUT=3600 # export NSR_HYPERVISOR_QUERY_REQUEST_TIMEOUT ## NW server interval to change VMware inventory (nsrvim) frequency (Default is 15 minutes) ## NOTE only supported in 19.10.0.0 and later # NSRVIM_TIME_INTERVAL=60 # Interval is in minutes and can be set between 15 and 60. # export NSRVIM_TIME_INTERVAL # Confirmation parameters added to end of script - comment or modify as desired. echo "### $(date +%FT%T) - Starting with nsrrc parameters ###" >> /nsr/nsrrc.log sysctl -a >> /nsr/nsrrc.log env >> /nsr/nsrrc.log ulimit -a >> /nsr/nsrrc.log
Windows:由于 /nsr/nsrrc 文件不存在,必须使用批处理文件运行更改,例如 nsrrc.bat 或其他部署方法。此处提供了命令,其中存在命令驱动的选项。这些更改是全局性的,不需要重复运行。就像 Linux 的 nsrrc 文件中,默认的标准条目未注释,非标准或间接选项注释。使用以下命令更改设置的可用性: REM 相关行上的前缀。根据部署文件的位置,剪裁文件使其与 NetWorker 客户端、节点或服务器相关。进行更改后,将需要重新启动服务。
REM ### WINDOWS - For all NetWorker hosts - Clients, Nodes and Server REM # TCP window size tuning - greater throughput / Data Domain REM reg add HKEY_LOCAL_MACHINE\SYSTEM\CurrentControlSet\Services\AFD\Parameters /v DefaultSendWindow /t REG_DWORD /d 262144 /f REM reg add HKEY_LOCAL_MACHINE\SYSTEM\CurrentControlSet\Services\AFD\Parameters /v DefaultReceiveWindow /t REG_DWORD /d 262144 /f REM reg add HKEY_LOCAL_MACHINE\SYSTEM\CurrentControlSet\Services\Tcpip\Parameters /v GlobalMaxTcpWindowSize /t REG_DWORD /d 262144 /f REM reg add HKEY_LOCAL_MACHINE\SYSTEM\CurrentControlSet\Services\Tcpip\Parameters /v TcpWindowSize /t REG_DWORD /d 262144 /f REM # Global keepalive registry settings - 270s to fall below common idle socket timer kills of 300s reg add HKEY_LOCAL_MACHINE\SYSTEM\CurrentControlSet\Services\Tcpip\Parameters /v KeepAliveTime /t REG_DWORD /d 270000 /f reg add HKEY_LOCAL_MACHINE\SYSTEM\CurrentControlSet\Services\Tcpip\Parameters /v KeepAliveInterval /t REG_DWORD /d 75000 /f reg add HKEY_LOCAL_MACHINE\SYSTEM\CurrentControlSet\Services\Tcpip\Parameters /v TcpMaxDataRetransmissions /t REG_DWORD /d 20 /f REM # Global NetWorker keepalive and connectivity variables setx /m NW_TCP_KEEPIDLE_SECS 270 setx /m NW_TCP_KEEPINTVL_SECS 75 setx /m NW_TCP_KEEPCNT 20 setx /m NSR_KEEPALIVE_WAIT 10 setx /m NSR_EXEC_MAX_AUTH_THREADS 50 REM setx /m NSR_SOCK_BUF_SIZE=65536 # (262144 for 10 Gb Eth NICs) REM ### For NetWorker Storage Nodes and Server REM # Standard TCP features - disable in case of disconnections REM netsh interface tcp set global rss=disabled REM netsh interface tcp set global autotuning=disabled REM netsh interface tcp set global ecncapability=disabled REM netsh interface tcp set global timestamps=default REM # Port range availability for TCP client callers netsh int ipv4 set dynamicport tcp start=10000 num=54000 netsh int ipv4 set dynamicport udp start=10000 num=54000 netsh int ipv6 set dynamicport tcp start=10000 num=54000 netsh int ipv6 set dynamicport udp start=10000 num=54000 REM # Global port maximum (deprecated) and TIME_WAIT window REM reg add HKEY_LOCAL_MACHINE\SYSTEM\CurrentControlSet\Services\Tcpip\Parameters /v MaxUserPort /t REG_DWORD /d 65535 /f reg add HKEY_LOCAL_MACHINE\SYSTEM\CurrentControlSet\Services\Tcpip\Parameters /v TcpTimedWaitDelay /t REG_DWORD /d 30 /f REM # Disable IPv6 if not required REM reg add HKEY_LOCAL_MACHINE\SYSTEM\CurrentControlSet\Services\Tcpip6\Parameters /v DisabledComponents /t REG_DWORD /d 0x000000ff /f REM ### For NetWorker Server only REM # Settings to increase device resilience for cloud operations or other potentially high-latency devices REM setx /m NSR_DEVOP_TIMEOUT 3600 REM setx /m NSR_DEVOP_POLLING_INTERVAL 600 REM setx /m NSR_DEVOP_INQUIRY_TIMEOUT 900 REM ### Settings for media database tuning REM setx /m NSR_TCP_READ_LONG_WAIT Y REM setx /m NSR_MAX_MEDIADB_RETRY 10 REM setx /m MDB_SQLITE_HEAP_MIN_ALLOC_SIZE 128 REM setx /m MMDB_SQLITE_CONFIGURE_MEMORY 1 REM setx /m MMDB_SQLITE_HEAP_SIZE 1073741824 REM setx /m MMDB_SQLITE_PAGE_COUNT 65536 REM setx /m MMDB_SQLITE_PAGECACHE_COUNT 65536 REM setx /m MMDB_SQLITE_TMP path_to_temp_dir REM ### NetWorker VMware Protection (NVP) Specific Tunables REM ## Increase the inventory (nsrvim) default timeout REM setx /m GST_VBA_TIMEOUT 7200 REM setx /m NSR_HYPERVISOR_QUERY_REQUEST_TIMEOUT 3600 REM ## NW server interval to change VMware inventory (nsrvim) frequency (Default is 15 minutes) REM ## NOTE only supported in 19.10.0.0 and later REM setx /m NSRVIM_TIME_INTERVAL 60 REM ## Interval is in minutes and can be set between 15 and 60.