7. Errors

This section discusses error and warning messages generated by the PHISH library and bait.py and bait.py tools. It also gives tips on debugging the operation of PHISH nets.

7.1 Debugguing PHISH nets 7.2 Error and warning messages from the PHISH library 7.3 Error messages from bait.py

Debugguing PHISH nets

A PHISH net can be difficult to debug because it may involve many independent running processes exchanging datums in a complex pattern, and rapid processing of large volumes of data. Here are some ideas that may be helpful in finding a bug if something goes wrong:

• As with any parallel program, running on as few processors as possible (that still exhibit the bug) simplifies debugging. For a PHISH net, each minnow is a process, but you may be able to reduce the minnow count via the school command in the PHISH input script.
• In principle, a PHISH net should behave similarly whether it is run entirely on a desktop machine, on a large parallel machine, or on a distributed network. Running all minnows on a single desktop machine is almost always the easiest mode to debug in.
• One or more slowdown minnows can be inserted in a PHISH net to slow down the rate at which datums are processed.
• The print minnow can receive the output of any minnow and print it to the screen. This is done by using the hook command in a PHISH input script. The output of a minnow can be connected to a print minnow in addition to other minnows. Also note that hooking a minnow's output port to another input port is optional, as discussed in this section. This means that you can often debug a PHISH net in stages by commenting out downstream minnows and then turning them on one at a time.
• Printf() or fprintf() statements can be added to a minnow's source code to print messages to the screen or a file when datums are received, processed, or sent. Doing this stage by stage, beginning with the first datums read or generated by the PHISH input script, is an effective way to verify that datums are formatted correctly, and are being sent and received as expected.

Error and warning messages from the PHISH library

When a minnow makes a call to the PHISH library, various error and warning conditions are checked for. If an error is encountered, a message in the following format is printed to stderr:

PHISH ERROR: Minnow executable ID school-id # global-id: message 

where "executable" is the name of executable minnow file (not the full path, just the filename), "school-id" is the ID of the minnow as specified in the PHISH input script, "global-id" is the global-ID of the minnow, and message is the error message. Each minnow has a global ID from 0 to Nglobal-1, where Nglobal is the total number of minnows in the net specified by the PHISH input script. This supplementary information is helpful in debugging which minnow generated the error message.

If a warning condition is detected, a similarly-formatted message is printed, with ERROR replaced by WARNING.

Note that by default, an error condition will cause an abort, shutting-down the entire PHISH net. However, it is possible for a minnow to explicitly cancel the abort. In this case, the PHISH library call that triggered the error will return a non-zero integer error code to indicate that an error occurred. See the documentation for phish_abort() for more information on how to cancel an abort.

The messages should be self-explanatory. See the doc page for the individual">>PHISH library, or the doc pages for PHISH library functions for relevant details. If necessary, the library source code in src/phish-common.cpp or src/phish-mpi.cpp or src/phish-zmq.cpp can be searched for the message text.

Error messages from bait.py

The bait.py tool is used to process PHISH input scripts and optionally run them. Any errors that bait.py encounters in the input script generate error messages that should be self-explanatory.

See the doc page for the individual">>bait.py tool, or the doc pages for PHISH input script commands for relevant details. If necessary, the Python source code in bait/bait.py can be searched for the message text.