Re: TRP Error Handling Spec Draft

[mwsachs@us.ibm.com]

This proposal is well thought-out and well described.

My concerns mainly relate to the KISS principle.  Is this a level of
complexity and detail that is really needed?  An alternative at the far
other end of the spectrum is to simply log everything and rely on
out-of-band communications for the messaging service instance which
received the message in error to notify the messaging service instance
which sent it.

When I was working with the IBM Poughkeepsie architecture team on the S/390
fiber optic channels, we accumulated a long list of errors which we wanted
reported in a response message.  The I/O engineers then informed us that
they were not going to analyze all this information.  Rather, they would
log the error condition and take the most drastic recovery action (called
connection recovery) whatever the error condition.  The recovery action is
simply to terminate the logical and physical connections between the
channel and I/O device and let software retry. The principle they were
operating on was KISS.  They said that ecovery function is the most
difficult to test and debug.

2.1.2 Types of errors

Line 21 (2nd bullet, 3rd subbullet):  Please add "of a previously sent
message" to the end of the line.

Following line 21:  What about a category of "an error in a document which
is reporting an error"?

2.1.3 When to Generate Error Messages

Line 30 (1st para.):   Please change "message in error" to "received
message".

Line 30 (1st para.):  Please change "a URL to send the message" to "an
address to send the message".  If the communication protocol is other than
HTTP, the address won't be a URL.

NOTE:  The IBM tpaML proposal covers only retries of transport timeout
errors and business-level errors.  Business-level errors are reported to
the sending application using <ExceptionResponse> and a corresponding
action at the other partner which receives the exception response.

NOTE:  It is not clear where the errors covered in the error-handling
proposal should be reported to.  They are conditions presumably caused by
the sending message service instance and detected by the receiving message
service instance.  We will need to define some kind of "catcher" for the
error messages and define what the catcher is supposed to do.

2.2 .1 Communication Protocol Errors

Line 55 (Editor's note): The KISS principle applies here.  I suggest
retrying communication-level errors at the communication level and
reporting all unrecoverable conditions to one place rather than defining a
unique recovery for each communication-level error.  Aside from simplifying
the implementation, it will avoid the need to track the other standards for
additional error codes.  I speculate that the function to which these
conditions are reported will simply log them and process all of them the
same way.

2.2.2.1  Identifying the Error Reporting Location

1st paragraph, line 63:  Please change "a URL" to "a communication
address".  The communication protocol will not necessarily be HTTP.

Line 65, by reference:  The TPAId is always in the ebXML message header

Line 66, by value:  Please replace "the URL" by "the communication
address".

Line 66, by value:  A header error will make the value of such a tag
untrustworthy.  I suggest that we limit ourselves to "by reference" and
"implicitly".

Line 74-76 ("Even if the message in error..."): Determining the error
reporting location is not a problem with either "by reference" or
"implicitly".

2.2.2.2  Error Reporting Location is Identified

Line 80-82 (Editor note): For communication errors, replacing an
acknowledgment message by an error message is OK (example:  HTTP). If the
message is sent at the messaging service level, it is not clear where the
error message should go.  Possibilities are:

   All the way to the application (e.g. the exception response message in
   tpaML).
   Some messaging service management function that we haven't yet
   considered.  I believe that this is the preferable approach for errors
   in the header and envelope.

2.2.2.3  Error Reporting Location not Identified

Line 84-86 (1st para.):  This is reporting the error at the transport
level.

Line 87, editor's note:  Specifying rules to apply to each transport
requires defining the layer of function that invokes the transport
protocol.

Lines 90-91 (If a suitable return address is not identified):  This is my
preferred approach for error processing in general.  If we define a
messaging service management function, then the error information could be
sent to the management function at the sender and logged there, as well as
being logged at the recipient messaging service management function.

2.2.3 Transient Errors

Line 92 (title):  Please change the title to "Recoverable Errors"  and use
the term "Recoverable"  instead of "transient" throughout.  "Recoverable"
means that retry is recommended.  "Transient" means that the error did not
recur, which can only be known after the retry succeeds.

Lines 105-106 ("If the ebXML messaging Services is unable to respond..."):
If the messaging service is unable to respond within the timeout period, it
is probably unable to send the message which says it is unable to respond.

Line 107 (determining the timeout period):  The TPA is one place to specify
the timeouts.

2.3.2  ebXML Error Document

Lines 136-137 (Editor's note, list item 2): Since correcting the errors
will require human intervention, logging the error condition should be
sufficient.  As noted above, the sender needs to be notified of the error
condition.  This notification could be outside the scope of the
specification.  Alternatively, the error message structure defined here
could be used for that purpose as long as it is sent to a management
function at the sender and does not require executing a protocol to
evaluate the error message as part of the exchange of messages between the
parties.

Line 139-40 (list item 4):  I completely agree.

2.3.5 ebXML Error Document Type Definition

Lines 256-258 (Editor's note): Standard practice that I am familiar with is
to terminate the operation on the first error detected and report only that
error.

2.4.2  Non-XML Document ERrors

Lines 330 (MsgTooLarge), 332 (MIMEError), 333 (MimePartMiss), and 336
(MimePartUnx):  These conditions appear to be application-level errors.
These could be reported directly to the sending application using a
mechanism like <ExceptionResponse> as defined in the tpaML proposal.

Regards,
Marty
*************************************************************************************

Martin W. Sachs
IBM T. J. Watson Research Center
P. O. B. 704
Yorktown Hts, NY 10598
914-784-7287;  IBM tie line 863-7287
Notes address:  Martin W Sachs/Watson/IBM
Internet address:  mwsachs @ us.ibm.com
*************************************************************************************



David Burdett <david.burdett@commerceone.com> on 08/18/2000 08:17:19 PM

To:   "ebXML Transport (E-mail)" <ebxml-transport@lists.ebxml.org>
cc:
Subject:  TRP Error Handling Spec Draft



Folks

I attach some light reading for the weekend. Alternatively called, the TRP
Error Handling draft spec version 01 ;)

David
 <<ebXML TRP Error Handling draft 01.doc>>  <<ebXML TRP Error Handling
draft
01.pdf>>

Product Management, CommerceOne
4400 Rosewood Drive 3rd Fl, Bldg 4, Pleasanton, CA 94588, USA
Tel: +1 (925) 520 4422 (also voicemail); Pager: +1 (888) 936 9599
mailto:david.burdett@commerceone.com; Web: http://www.commerceone.com