e7ecd56bd2
such, the commit message back to the master SVN repository is fairly long. = ORTE Job-Level Output Messages = Add two new interfaces that should be used for all new code throughout the ORTE and OMPI layers (we already make the search-and-replace on the existing ORTE / OMPI layers): * orte_output(): (and corresponding friends ORTE_OUTPUT, orte_output_verbose, etc.) This function sends the output directly to the HNP for processing as part of a job-specific output channel. It supports all the same outputs as opal_output() (syslog, file, stdout, stderr), but for stdout/stderr, the output is sent to the HNP for processing and output. More on this below. * orte_show_help(): This function is a drop-in-replacement for opal_show_help(), with two differences in functionality: 1. the rendered text help message output is sent to the HNP for display (rather than outputting directly into the process' stderr stream) 1. the HNP detects duplicate help messages and does not display them (so that you don't see the same error message N times, once from each of your N MPI processes); instead, it counts "new" instances of the help message and displays a message every ~5 seconds when there are new ones ("I got X new copies of the help message...") opal_show_help and opal_output still exist, but they only output in the current process. The intent for the new orte_* functions is that they can apply job-level intelligence to the output. As such, we recommend that all new ORTE and OMPI code use the new orte_* functions, not thei opal_* functions. === New code === For ORTE and OMPI programmers, here's what you need to do differently in new code: * Do not include opal/util/show_help.h or opal/util/output.h. Instead, include orte/util/output.h (this one header file has declarations for both the orte_output() series of functions and orte_show_help()). * Effectively s/opal_output/orte_output/gi throughout your code. Note that orte_output_open() takes a slightly different argument list (as a way to pass data to the filtering stream -- see below), so you if explicitly call opal_output_open(), you'll need to slightly adapt to the new signature of orte_output_open(). * Literally s/opal_show_help/orte_show_help/. The function signature is identical. === Notes === * orte_output'ing to stream 0 will do similar to what opal_output'ing did, so leaving a hard-coded "0" as the first argument is safe. * For systems that do not use ORTE's RML or the HNP, the effect of orte_output_* and orte_show_help will be identical to their opal counterparts (the additional information passed to orte_output_open() will be lost!). Indeed, the orte_* functions simply become trivial wrappers to their opal_* counterparts. Note that we have not tested this; the code is simple but it is quite possible that we mucked something up. = Filter Framework = Messages sent view the new orte_* functions described above and messages output via the IOF on the HNP will now optionally be passed through a new "filter" framework before being output to stdout/stderr. The "filter" OPAL MCA framework is intended to allow preprocessing to messages before they are sent to their final destinations. The first component that was written in the filter framework was to create an XML stream, segregating all the messages into different XML tags, etc. This will allow 3rd party tools to read the stdout/stderr from the HNP and be able to know exactly what each text message is (e.g., a help message, another OMPI infrastructure message, stdout from the user process, stderr from the user process, etc.). Filtering is not active by default. Filter components must be specifically requested, such as: {{{ $ mpirun --mca filter xml ... }}} There can only be one filter component active. = New MCA Parameters = The new functionality described above introduces two new MCA parameters: * '''orte_base_help_aggregate''': Defaults to 1 (true), meaning that help messages will be aggregated, as described above. If set to 0, all help messages will be displayed, even if they are duplicates (i.e., the original behavior). * '''orte_base_show_output_recursions''': An MCA parameter to help debug one of the known issues, described below. It is likely that this MCA parameter will disappear before v1.3 final. = Known Issues = * The XML filter component is not complete. The current output from this component is preliminary and not real XML. A bit more work needs to be done to configure.m4 search for an appropriate XML library/link it in/use it at run time. * There are possible recursion loops in the orte_output() and orte_show_help() functions -- e.g., if RML send calls orte_output() or orte_show_help(). We have some ideas how to fix these, but figured that it was ok to commit before feature freeze with known issues. The code currently contains sub-optimal workarounds so that this will not be a problem, but it would be good to actually solve the problem rather than have hackish workarounds before v1.3 final. This commit was SVN r18434. |
||
---|---|---|
.. | ||
base | ||
null | ||
proxy | ||
svc | ||
iof.h | ||
Makefile.am | ||
README.txt |
Some notes from Jeff/Rolf while trying to figure out how IOF works... 1. E-mail from Rolf->Jeff with some pretty ASCII art 2. Notes from Jeff about problems we'll have when/if we ever try to use IOF more creatively. =============================================================================== From: Rolf.Vandevaart@Sun.COM Subject: Picture of IOF side of things. Date: June 7, 2007 10:43:42 AM EDT To: jsquyres@cisco.com Not sure if this will come out for you, but here is an ASCII represantation of what the HNP looks like after IOF has been wired up. VIEW OF DATA STRUCTURES ON IOF SVC SIDE KEY: ORTE_IOF_SOURCE=0 ORTE_IOF_SINK=1 After the job starts up, this is how things look on the HNP side of things. ENDPOINTS (orte_iof_base_endpoint_t) mode origin tag fd seq ack src_frags sink_frags notes ============================================================================================= 1 0,0,0 1 1 0 0 0 0 pull() call from rmgr<- ============================================================================================= | 1 0,0,0 2 2 0 0 0 0 pull() call from rmgr<-|--- ============================================================================================= | | 0 0,0,0 0 0 0 0 0 0 push() call from rmgr | | ============================================================================================= | | | | | | SUBSCRIBERS (orte_iof_svc_sub_t) | | ORIGIN TARGET (list) | | name mask tag name mask tag endpoint forward has_been_acked last_ack_forwarded | | ====================================================================================== | | 0,1,0 2 1 0,0,0 2 1 0 | | ptr-------------------------------------------------| | ====================================================================================== | 0,1,0 2 2 0,0,0 2 2 0 | ptr-----------------------------------------------------| ====================================================================================== 0,0,0 ff 0 0,1,0 2 0 NULL 1 ----------------------------------- | ====================================================================================== | | | PUBLISHED (orte_iof_svc_pub_t) | name proxy mask tag endpoint | ========================================================================================= | 0,1,0 0,0,1 ff 0 NULL <------------------ ========================================================================================= FORWARD (orte_iof_svc_fwd_t) This structure is just a connection from a subscriber to publisher. I have omitted it in the drawings. However, it is worth pointing out the structure as I am not clear on why we have the table. struct orte_iof_svc_fwd_t { opal_list_item_t super; orte_iof_svc_pub_t* fwd_pub; opal_hash_table_t fwd_seq_hash; }; Note: This first subscriber says that it will receive from any process in the job. Note that the jobid=1 and the mask=2. So, we expect this to collect the stdout from any of the ranks. Obviously the second subscriber says the same thing but for stderr. The third subscriber is for receving data from stdin and sending it out to rank 0 of the job. Notice the mask=ff which means compare jobid,vpid when addressing where the data goes. The first endpoint is created by a call to pull by the rmgr. After the endpoint is created, a subscription is created as well. Then, the subscription is tied to the endpoint. For the stdin creation, we first create the subscription, and then the endpoint. In that way, the endpoint is not found and does not get tied to the subscription. Hmmm, this I do not really understand. APPENDIX A These are the defines that go with the mask. #define ORTE_NS_CMP_NONE 0x00 #define ORTE_NS_CMP_JOBID 0x02 #define ORTE_NS_CMP_VPID 0x04 #define ORTE_NS_CMP_ALL 0Xff When we get a HDR_MSG, we call orte_iof_svc_proxy_msg() APPENDIX B There are two dbx files that help get to where we want to get for seeing how things work. start.x : Run this first to get initial breakpoint. Needs this so we can set additional breakpoints. This also has some very helpful aliases for looking at the structures shown above. follow.x : Run this second to set initial breakpoints and setup some useful aliases. =============================================================================== Random notes from Jeff: - Many issues may not come up because we only have single subscribers; I'm sure new things will come up. Examples: - What happens if all subscribers to a stream disconnect, and then a new subscriber connects? I'm guessing the ACKs will be all screwed up and we'll never end up reading from that fd again (because it will likely be stalled because not enough acks have been received, and therefore it removed itself from the event engine). - If all subscribers disconnect from a stdin/SINK, chances are that we'll lose the last frag that was sent before the disconnect. I.e., if there was a frag in flight when the disconnect was received, that frag is effectively lost. So if someone reconnects to the stdin stream later, it won't start reading exactly where the first subscriber left off. We need to define what is *supposed* to happen here... - odls default: make handling of vpid 0 uniform between setup and takedown -- some kind of global variable, perhaps? (not multi-proc / thread safe) - odls default: currently, we publish stdin (if relevant), stdout, and stderr (note that only the stdin publish message gets sent to svc; the publish for SOURCEs stdout/stderr is not actually sent to the svc because all SOURCE frags are sent to the svc automatically). But we only unpublish stdout. I think we should either: - publish stdin, stdout, stderr, and unpublish stdin, stdout, stderr or - publish stdin, and unpublish stdin I.e., make the code symmetric. Note, however, that unpublish for STDOUT/STDERR are sent to the svc (whereas publish for STDOUT/STDERR are not). So if we unpublish stdout/stderr, we'll be creating a storm to the svc upon shutdown (i.e,. scalability problems). :-( - for scalability, we want to be able to change the proxy to *not* unconfitionally send everything to svc. But this has the problem that if we do this, then we have to send the publish request to the svc (which we don't today since everything just automatically goes to svc). But then in the common case (where vpid!=0 has no stdout/stderr), we're flooding svc with N publish requests from all the vpids, simply creating a different scalability problem (during startup). - random q: are the proxy publish requests not sent back to svc because it prevents a storm of publish requests during startup? I.e., this was intentional to give better scalability? Could be; but it still seems weird... Perhaps a better scheme would be to have the IOF *assume* that the stdin/stdout/stderr are all published upon startup (or be told by a single control message; perhaps in the app context?) and further *assume* that they are all unpublished when the job completes. Putting this info in the app context (for example) might jive with a more capable orterun that allows flexible stdin/stdout/stderr mapping (think: mpirun --screen ...). mpirun makes the decision about how to wire up stdin/stdout/stderr and includes it in the app context (or whatever). This is given to the svc who then creates publications as relevant. Upon job completion, all publications/subscriptions related to that job are destroyed.