README file for ssh-spawner =========================== @ TOC: @ @ Section: Overview @ @ Section: Basic Usage @ @ Section: Process Layout @ @ Section: Build-time Configuration @ @ Section: Runtime Configuration @ @ Section: Troubleshooting Connection Problems @ @ Section: Overview @ This document describes ssh-spawner, a component of the GASNet runtime library which allows applications using many GASNet conduits to utilize ssh (or rsh) to perform job launch. If installing GASNet on system which allows remote shell connections to the compute nodes, this may be a lower-overhead mechanism for job launch than use of the default mpi-spawner. This is also the default spawner when MPI support is not present (or is disabled). @ Section: Basic Usage @ + Usage summary (option 1): Many languages and libraries built over GASNet provide their own commands for job launch. These should be used instead of GASNet's whenever possible. They typically wrap the mechanisms described below, while providing additional options specific to the language or library. The remaining options are documented here mainly for those who are implementing such a wrapper. + Usage summary (option 2): Conduits which support ssh-spawner each include a spawner utility named for the conduit: gasnetrun_[conduit] -n [options] [--] prog [program args] options: -n number of processes to run (required) -N number of nodes to run on (not supported by all MPIs) -E list of environment vars to propagate -v be verbose about what is happening -t test only, don't execute anything (implies -v) -k keep any temporary files created (implies -v) -spawner=(ssh|mpi|pmi) force use of a specific spawner (if available) The ssh-spawner described in this README is used if selected by one of the following three mechanisms, in order from greatest to least precedence: + Passing -spawner=ssh to the gasnetrun_[conduit] utility + Setting GASNET_[CONDUIT]_SPAWNER=ssh in the environment + If ssh-spawner was established as the default at configure time (see Build-time Configuration, below). @ Section: Process Layout @ The ssh-spawner will layout processes in a "balanced" distribution and "blocked" order on a list of hosts (such as obtained from the GASNET_SSH_SERVERS environment variable). For P processes and N hosts, "balanced" distribution places ceil(P/N) processes on the first (P%N) hosts and floor(P/N) on the remainder. For P divisible by N, this yields P/N processes on every host, while for other all cases the last (N-P%N) hosts each have one fewer than the others. The "blocked" order means the processes on each host are numbered consecutively, with the first host holding processes starting from rank 0, the second holding processes starting from rank ceil(P/N), etc. By default the GASNET_SSH_SERVERS environment variable (or equivalent) is subject to de-duplication. However, by disabling this behavior (see GASNET_SSH_KEEPDUP environment variable below) one can exercise additional control over the placement of process though duplication of hostnames. For instance, with P=8, GASNET_SSH_SERVERS="node1 node2 node1 node2" and GASNET_SSH_KEEPDUP=1, the host "node1" will hold processes with ranks 0, 1, 4 and 5, rather than 0, 1, 2 and 3 as would be the case without setting GASNET_SSH_KEEPDUP. In the extreme case, populating GASNET_SSH_SERVERS with P entries allows for precise control over placement of every process, when de-duplication is disabled via GASNET_SSH_KEEPDUP=1. @ Section: Build-time Configuration @ The ssh-spawner offers the following configure-time options: --with-[conduit]-spawner=ssh Conduits which support ssh-spawner each accept a configure option of this form to set the default spawner used by the corresponding gasnetrun_[conduit] utility, as described in the "Basic Usage" section above. If this option is not used, the "mpi" is the default if MPI support is found at configure time, and "ssh" otherwise. However, one may pass this option to make ssh-spawner the default for the corresponding conduit. --disable-pdeathsig (auto-detect by default) On Linux, it is possible to request a signal be delivered to a process when its parent process dies. This can be used by the ssh-based spawner to reduce the possibility of orphan (run away) processes in certain abnormal termination scenarios. Because there are 2.4.x versions of Linux where use of this option can lock up the machine (as the result of a kernel bug), this option is disabled for kernels prior to 2.6.0, and can also be explicitly disabled at configure time. --with-ssh-{nodefile,cmd,options}= These control the default values used when the corresponding environment variables are not set. These environment variables are documented below. --with-ssh-out-degree= This establishes a default value for the GASNET_SSH_OUT_DEGREE environment variable, documented below. If your system does not permit remote shell connections among compute nodes, then you should configure using --with-ssh-out-degree=0 to ensure that ssh-spawner only attempts to make connections out-bound from the "master node". Be aware, however, that this may severely limit the size of jobs that one can launch. @ Section: Runtime Configuration @ Environment Variables: + A list of hosts is specified using one of the GASNET_SSH_NODEFILE, GASNET_SSH_SERVERS, or GASNET_NODEFILE environment variables (with precedence in that order). If set, variables GASNET_SSH_NODEFILE or GASNET_NODEFILE specify a file with one hostname per line. Blank lines and comment lines (using '#') are ignored. If set, the variable GASNET_SSH_SERVERS itself contains a list of hostnames, delimited by commas or whitespace. For sites using a static hosts file, a default value for the GASNET_SSH_NODEFILE variable may be set at configure time using the option --with-ssh-nodefile=. HOWEVER, if this is done then *only* setting this variable manually can override its default setting (since it has the highest precedence). Note that if starting a job via upcrun or tirun, these variables may be set for you from other sources. The following environment variables set by supported batch systems are also recognized if the GASNET_* variables are not set: PBS: PBS_NODEFILE LSF: LSB_HOSTS SGE: PE_HOSTFILE SLURM: Use `scontrol show hostname` if SLURM_JOB_ID is set + The environment variable GASNET_SSH_CMD can be set to specify a specific remote shell (perhaps rsh), without arguments (see below). If the value does not begin with "/" then $PATH will be searched to resolve a full path. The default value is "ssh", unless an other value has been configured using --with-ssh-cmd=. + The environment variable GASNET_SSH_OPTIONS can be set to specify options that will precede the hostname in the commands used to spawn jobs. One example, for OpenSSH, would be GASNET_SSH_OPTIONS="-o 'StrictHostKeyChecking no'" The parsing of the value follows the same rules for quotes (both single and double) and backslash as most shells. A default value may be configured using --with-ssh-options=. + The environment variable GASNET_SSH_OUT_DEGREE can be used to limit the number of out-going ssh connections from any given host. The value 0 means no limit is imposed. The default value is 32 unless an alternate value was set at configure time using --with-ssh-out-degree=. + The environment variable GASNET_SSH_REMOTE_PATH can be set to specify the working directory (defaults to current). + Users of OpenSSH should NOT add "-f" to GASNET_SSH_OPTIONS. Doing so causes the spawner to mistakenly believe that a process which it has spawned has exited. However, if agent forwarding or X11 forwarding are normally enabled in your configuration, "-a" and "-x" can be used with OpenSSH to disable them and speed the connection process (except where the agent forwarding is needed for authorization). + The environment variable GASNET_MASTERIP can be used to specify the exact IP address which the compute nodes should use to connect to the master (spawning) node. By default the master node will pass the result of gethostname() to the worker nodes, which will then resolve that to an IP address using gethostbynname(). + The environment variable GASNET_SSH_KEEPDUP controls the treatment of duplicate entries in GASNET_SSH_NODEFILE, GASNET_SSH_SERVERS and other sources of host lists. By default, GASNET_SSH_KEEPDUP=0 and duplicates are removed, keeping only the first instance of any given name. Note that this is based on string comparison only and does not consider multiple valid names for the same host to be duplicates. Setting GASNET_SSH_KEEPDUP=1 preserves the host list without any changes. @ Section: Troubleshooting Connection Problems @ For the following, the term "compute node" means one of the hosts given by GASNET_SSH_NODEFILE, GASNET_SSH_SERVERS, or obtained from a variable specific to the batch system. These are nodes which will run application processes. The term "master node" means the node from which the job was spawned. The master node may also be a compute nodes, but this is not required. The term "resolve" means translation of a hostname to an IP address using gethostbynname(). + The ssh (or rsh) at your site must be configured to allow logins from the master node to compute nodes. It is also strongly recommended that logins between compute nodes be permitted (see the description of the --with-ssh-out-degree configure option for the alternative). These must be achieved without user interaction such as entering a password or accepting hostkeys. For OpenSSH users, the following options are used automatically -o 'StrictHostKeyChecking no' -o 'BatchMode yes' which should ensure that ssh does not try to prompt the user. There is no need to specify these in the GASNET_SSH_OPTIONS environment variable. + Any firewall or port filtering must allow the ssh/rsh connections described above, plus TCP connections on an "untrusted port" (ports with numbers over 1024) from a compute node to the master node (and among compute nodes unless the out-degree has been set to 0). + Setting of the GASNET_MASTERIP environment variable trivially provides "resolution" of the master node's hostname. Otherwise: - The master node and all compute nodes must be able to resolve the hostname of the master node. + Providing only numeric IPs for compute nodes (in GASNET_SSH_SERVERS, the file named by GASNET_SSH_NODEFILE, or other source specific to the batch system) trivially provides "resolution" of the hostnames of all compute nodes. Otherwise: - The master node must be able to resolve the hostnames of all compute nodes. - Unless the out-degree has been set to zero, compute nodes must able to resolve the hostnames of the other compute nodes.