1
1
openmpi/orte/mca/iof/iof.h
Ralph Castain 9613b3176c Effectively revert the orte_output system and return to direct use of opal_output at all levels. Retain the orte_show_help subsystem to allow aggregation of show_help messages at the HNP.
After much work by Jeff and myself, and quite a lot of discussion, it has become clear that we simply cannot resolve the infinite loops caused by RML-involved subsystems calling orte_output. The original rationale for the change to orte_output has also been reduced by shifting the output of XML-formatted vs human readable messages to an alternative approach.

I have globally replaced the orte_output/ORTE_OUTPUT calls in the code base, as well as the corresponding .h file name. I have test compiled and run this on the various environments within my reach, so hopefully this will prove minimally disruptive.

This commit was SVN r18619.
2008-06-09 14:53:58 +00:00

365 строки
13 KiB
C

/*
* Copyright (c) 2004-2007 The Trustees of Indiana University and Indiana
* University Research and Technology
* Corporation. All rights reserved.
* Copyright (c) 2004-2005 The University of Tennessee and The University
* of Tennessee Research Foundation. All rights
* reserved.
* Copyright (c) 2004-2005 High Performance Computing Center Stuttgart,
* University of Stuttgart. All rights reserved.
* Copyright (c) 2004-2005 The Regents of the University of California.
* All rights reserved.
* Copyright (c) 2007-2008 Cisco, Inc. All rights reserved.
* $COPYRIGHT$
*
* Additional copyrights may follow
*
* $HEADER$
*/
/**
* @file
*
* I/O Forwarding Service
* The I/O forwarding service (IOF) is used to push file descriptor
* streams between ORTE processes. It is currently primarily used to
* push stdin, stdout, and stderr between ORTE processes, but can be
* used with any file descriptor stream.
*
* In practice, the IOF acts as a multiplexor between local file
* descriptors and the RML; the RML relays information from local file
* descriptors to remote file descriptors. Note that the IOF allows
* many-to-one mappings; SOURCE streams can be directed to multiple
* destinations and SINK streams can receive input from multiple
* sources.
*
* The design is fairly simple: streams are designated as either
* ORTE_IOF_SOURCEs or ORTE_IOF_SINKs. SOURCE streams provide content
* that is pushed elsewhere. SINK streams accept content that
* originated from elsewhere. In short, we read from SOURCEs and we
* write to SINKs.
*
* Streams are identified by ORTE process name (to include wildecards,
* such as "all processes in ORTE job X") and tag. There are
* currently 4 predefined tags, although any integer value is
* sufficient:
*
* - ORTE_IOF_ANY (value -1): any stream will match
* - ORTE_IOF_STDIN (value 0): recommended for file descriptor 0, or
* wherever the standard input is currently tied.
* - ORTE_IOF_STDOUT (value 1): recommended for file descriptor 1, or
* wherever the standard output is currently tied.
* - ORTE_IOF_STDERR (value 2): recommended for file descriptor 2, or
* wherever the standard error is currently tied.
* - ORTE_IOF_INTERNAL (value 3): recommended for "internal" messages
* from the infrastructure, just to differentiate them from user job
* stdout/stderr
*
* Note that since streams are identified by ORTE process name, the
* caller has no idea whether the stream is on the local node or a
* remote node -- it's just a stream.
*
* IOF components are selected on a "one of many" basis, meaning that
* only one IOF component will be selected for a given process.
* Details for the various components are given in their source code
* bases.
*
* The following basic actions are supported in IOF:
*
* publish: File descriptors are "published" as a stream as a
* mechanism to make them available to other processes. For example,
* if a stdout descriptor is available from process X, then process X
* needs to publish it (and make it a stream) in order to make that
* stdout stream available to any other process.
* --> today, this isn't necessarily true for the proxy because
* everything is atuomatically sent to the svc. But the proxy
* should be fixed someday to make this definition consistent.
*
* unpublish: The opposite of publish; when a stream is unpublished,
* the content from that file desciptor is no longer available to
* other processes.
*
* push: Tie together a local file descriptor (*not* a stream!) that
* should be treated as a source to a stream that should be treated as
* a SINK. Subsequent input that appears on the file descriptor will
* automatically be pushed to the SINK stream. There is currently no
* way to stop a push; once it starts, it runs until an EOF is
* received on the file descriptor or the target stream is
* unpublished.
*
* pull: Tie together a local file descriptor (*not* a stream!) that
* should be treated as a sink to a stream that should be treated as a
* SOURCE. Subsequent input that appears via the stream will
* automatically be sent to the target file descriptor. There is
* currently no way to stop a pull; once it starts, it runs until an
* EOF is receives on the file descriptor or the source stream is
* unpublished.
*
* subscribe: Setup a callback function that is invoked whenever a
* fragment from a matching stream arrives. This can be used to
* post-process fragment information, such as prepending a prefix to
* stdout data before outputting it to the user's display in order to
* identify the source process.
*
* unsubscribe: Remove a callback that was previously setup via the
* subscribe action.
*
* flush: Block until all pending data has been written down local
* file descriptors and/or completed sending across the OOB to remote
* process targets.
*
* Two terms that are used in the IOF interface are "origin" and
* "target" indicating the process where data started and where it is
* going. These terms are used to distinguish IOF component
* implementation details because data does not necessarily only from
* the SOURCE process to the SINK process. In practice, data can flow
* from a SINK to a SOURCE (e.g., an ACK), or be routed through a
* proxy. So the "origin" and "target" processes are those where the
* data started and will terminate, respectively, regardless of the
* designation of the originating process (as the SOURCE, SINK, or
* proxy) and the destination process (as the SOURCE, SINK, or proxy).
*
* Additionally, the "proxy" is as it is described above: it may be
* the origin or target itself, or it may be an intermediary acting on
* behalf of the origin or target.
*
* Examples:
*
* 1. mpirun -np 1 hostname
* Assume that orteds and an HNP are used. An orted will be
* launched on the same node as "hostname". It will act as a proxy
* for the hostname process' stdin, stdout, and stderr. Data read
* by the orted from the hostname process stdout will be sent to
* mpirun. In this case, the hostname process is the origin,
* mpirun is the target, and the orted is the proxy.
*
* 2. mpirun -np 1 read_stdin < input_filename
* Assume that orteds and an HNP are used. As with #1, an orted
* will proxy the stdin, stdout, and stderr of the read_stdin
* process. When mpirun reads data on its stdin, it will forward
* it to the orted to write down the pipe to the read_stdin
* process. In this case, mpirun is both the origin process *and*
* proxy, and read_stdin is the target.
*/
#ifndef ORTE_IOF_H
#define ORTE_IOF_H
#include "orte_config.h"
#include "orte/types.h"
#include "opal/class/opal_list.h"
#include "opal/mca/mca.h"
#include "opal/mca/crs/crs.h"
#include "opal/mca/crs/base/base.h"
#include "orte/util/name_fns.h"
BEGIN_C_DECLS
/* Predefined tag values */
enum {
ORTE_IOF_ANY = -1,
ORTE_IOF_STDIN = 0,
ORTE_IOF_STDOUT = 1,
ORTE_IOF_STDERR = 2,
ORTE_IOF_INTERNAL = 3
};
typedef int orte_iof_base_tag_t;
/* endpoint mode */
enum {
ORTE_IOF_SOURCE = 0,
ORTE_IOF_SINK
};
typedef int orte_iof_base_mode_t;
/**
* Publish a local file descriptor as an endpoint that is logically
* associated with the specified origin process name. The file
* descriptor may be local to this process (in which case the origin
* process name is this process' name), or it may be a pipe to another
* process (i.e., this process is acting as a proxy for another
* process -- typically the case for stdin, stdout, stderr).
*
* @param origin Origin process name associated with the endpoint (not
* the proxy process).
* @param mode Is the endpoint an input or output (SOURCE or SINK)
* @param tag The logical tag associated with this file descriptor.
* @param fd Local file descriptor
*
*/
typedef int (*orte_iof_base_publish_fn_t)(
const orte_process_name_t* origin,
orte_iof_base_mode_t mode,
orte_iof_base_tag_t tag,
int fd
);
/**
* Remove all endpoints matching the specified origin process name,
* mask and tag values.
*
* @param name Origin process name associated with the endpoint.
* @param mask A mask indicating the set of processes to unpublish.
* @param tag The endpoint tag.
*
*/
typedef int (*orte_iof_base_unpublish_fn_t)(
const orte_process_name_t* origin,
orte_ns_cmp_bitmask_t mask,
orte_iof_base_tag_t tag
);
/**
* Callback function for subscriptions (see orte_iof_base_subscribe_fn_t).
*/
typedef void (*orte_iof_base_callback_fn_t)(
orte_process_name_t* origin_name,
orte_iof_base_tag_t orign_tag,
void *cbdata,
const unsigned char* data,
size_t count
);
/**
* Subscribe to receive a callback on receipt of data from a specified
* set of origin peers.
*
* This function is a general purpose utility for getting data from a
* stream; the incoming fragment is delivered to the callback in a
* buffer. You can do whatever you want with the buffer when you get
* the callback (e.g., buffer it, call syslog, ...etc.).
*
* Note that the orte_iof_base_pull_fn_t is a customized common-case
* version of this function; it always takes incoming fragments from a
* stream and writes them down an fd.
*/
typedef int (*orte_iof_base_subscribe_fn_t)(
const orte_process_name_t* origin_name,
orte_ns_cmp_bitmask_t origin_mask,
orte_iof_base_tag_t origin_tag,
orte_iof_base_callback_fn_t cb,
void* cbdata
);
/**
* Delete a subscription created by orte_iof_base_subscribe_fn_t.
*/
typedef int (*orte_iof_base_unsubscribe_fn_t)(
const orte_process_name_t* origin_name,
orte_ns_cmp_bitmask_t src_mask,
orte_iof_base_tag_t src_tag
);
/**
* Explicitly push data from the specified input file descriptor to
* the indicated set of SINK peers.
*
* This function is a shortcut for publishing a SOURCE stream and
* tying that stream to an fd that is providing data to be sent across
* the stream (e.g., read from stdin and push it out to a stream).
* Any data that appears on the fd will automatically be read and sent
* across the stream.
*
* @param sink_name Name used to qualify set of target peers.
* @param sink_mask Mask that specified how name is interpreted.
* @param sink_tag Match a specific peer endpoint.
* @param fd Local file descriptor for input.
*/
typedef int (*orte_iof_base_push_fn_t)(
const orte_process_name_t* sink_name,
orte_ns_cmp_bitmask_t sink_mask,
orte_iof_base_tag_t sink_tag,
int fd
);
/**
* Explicitly pull data from the specified set of SOURCE peers and
* dump to the indicated output file descriptor.
*
* This function is a shortcut for subscribing to a SOURCE stream and
* tying that stream to an fd that will consume the data received from
* the stream (i.e., get a fragment from a stream and write it down an
* fd). Any fragments that arrive on the stream will automatically be
* written down the fd.
*
* @param source_name Name used to qualify set of origin peers.
* @param source_mask Mask that specified how name is interpreted.
* @param source_tag Match a specific peer endpoint.
* @param fd Local file descriptor for output.
*/
typedef int (*orte_iof_base_pull_fn_t)(
const orte_process_name_t* source_name,
orte_ns_cmp_bitmask_t source_mask,
orte_iof_base_tag_t source_tag,
int fd
);
/**
* Flush all output and block until output is delivered.
*/
typedef int (*orte_iof_base_flush_fn_t)(void);
/**
* Shut down an IOF module
*/
typedef int (*orte_iof_base_finalize_fn_t)(void);
/**
* FT Event Notification
*/
typedef int (*orte_iof_base_ft_event_fn_t)(int state);
/**
* IOF module.
*/
struct orte_iof_base_module_1_0_0_t {
orte_iof_base_publish_fn_t iof_publish;
orte_iof_base_unpublish_fn_t iof_unpublish;
orte_iof_base_subscribe_fn_t iof_subscribe;
orte_iof_base_unsubscribe_fn_t iof_unsubscribe;
orte_iof_base_push_fn_t iof_push;
orte_iof_base_pull_fn_t iof_pull;
orte_iof_base_flush_fn_t iof_flush;
orte_iof_base_finalize_fn_t iof_finalize;
orte_iof_base_ft_event_fn_t ft_event;
};
typedef struct orte_iof_base_module_1_0_0_t orte_iof_base_module_1_0_0_t;
typedef orte_iof_base_module_1_0_0_t orte_iof_base_module_t;
ORTE_DECLSPEC extern orte_iof_base_module_t orte_iof;
/**
* IOF component init function. Contains component version
* information and component open/close/init functions.
*/
typedef orte_iof_base_module_t* (*orte_iof_base_component_init_fn_t)(
int *priority,
bool *allow_user_threads,
bool *have_hidden_threads
);
struct orte_iof_base_component_1_0_0_t {
mca_base_component_t iof_version;
mca_base_component_data_1_0_0_t iof_data;
orte_iof_base_component_init_fn_t iof_init;
};
typedef struct orte_iof_base_component_1_0_0_t orte_iof_base_component_1_0_0_t;
typedef struct orte_iof_base_component_1_0_0_t orte_iof_base_component_t;
END_C_DECLS
/*
* Macro for use in components that are of type iof v1.0.0
*/
#define ORTE_IOF_BASE_VERSION_1_0_0 \
/* iof v1.0 is chained to MCA v1.0 */ \
MCA_BASE_VERSION_1_0_0, \
/* iof v1.0 */ \
"iof", 1, 0, 0
#endif /* ORTE_IOF_H */