Re: Bug in HTTP API's http_req() procedure - leaving connections open

[Charles Wilt <charles.wilt@xxxxxxxxx>]

Just a follow up....

It appears that other programs running in the job are leaving
connections/files open (by design (mostly?)) and once we happen to get a
socket# higher than 224 we start seeing issues in HTTP API; and the
response file and/or connection gets left open when HTTP API crashes.

The Java JVM for instance has 180+ open file descriptors, mostly to various
JAR files.

When we run our new HTTP API process all by itself in a job, we don't see
any issues.

Charles



On Fri, Jan 11, 2019 at 11:27 AM Charles Wilt <charles.wilt@xxxxxxxxx>
wrote:

> Since the FTPAPI list is having problems, thought I'd post here...
>
> Running v1.33 of HTTP API, and noticing connections sometimes being left
> open by the HTTPS calls that use the http_req() procedure.
>
> Noticeable only because it's happening in a set of long running job making
> 50,000+ calls...
>
> Problem manifested as "out of range" error
> "One of the following has occurred in RPG procedure COMMTCP_FD in program
> LIBHTTP/HTTPAPIR4: - A numeric length or start position is less than 1 or
> too large for the string operation.describe her in this message"
>
> as described here...
> http://www.scottklement.com/archives/ftpapi/201504/msg00093.html
>
> It seems we're a few versions behind...
>
> Changes to version 1.39 from 1.38:
>
>   - Fixed ALGD0200 declaration in ENCRYPTR4 -- though, this should
>
>     not impact current users. (Thomas Raddatz)
>
>   - Correct international handling of $Version string in Cookie
>
>     header routines.
>
> Changes to version 1.38 from 1.37:
>
>   - Added memory alloc/dealloc diagnostic routines
>
>   - Fixed memory leaks -- comm drivers were not cleaned up
>
>   - fixed mismatches with regular alloc vs xalloc
>
> Changes to version 1.37 from 1.36:
>
>   - Fix problem where debugLevel not propagated properly
>
>   - Added buffering to COMMTCPR4 to improve LineRead performance
>
>   - Added buffering to COMMSSLR4 to improve LineRead performance
>
> Changes to version 1.36 from 1.35:
>
>   - Fix problem where http_persist_close() would get a pointer error
>
>     if http_persist_open() failed during SSL handshake.
>
>
> The changes in 1.36 and 1.38 seem most likely to be related.  Trying to
> get the latest version put on to test.
>
> But I'm hoping somebody here as seen the issue and can confirm it's fixed
> in a later version when I push management for the upgrade.
>
> Thanks!
> Charles