initialise_event_info_struct: initialise failed!

[roxkamx]

Hi,

I have a project where I attach 32 devices grouped in 2 physical hubs ( 16 each ) to a Centos7 server.
With 16 devices my application works just fine.

When I start the application with both hubs attached ( i.e. 32 devices attached ) at the 24th device my application stops with the following error:

initialise_event_info_struct: initialise failed!

This message looks to be from RXTX ( as found on http://rxtx.qbang.org/pub/rxtx/rxtx-2.1-7r2/src/SerialImp.c )

Any suggestions how to overcome this issue? is there any limitation?
The same thing happened on 2 other different PCs, so I would exclude for now the USB driver.

Thanks

[madhephaestus]

try 5.0.2

[madhephaestus]

unfortunately as i try to understand how this is happening, i have run into the 'goto' problem. Fail can be reached by 2 ways, one is id this statement returns true:

if (eis->fd < FD_SETSIZE && eis->fd > 0) 

and the other

if(eis->send_event == NULL)

[madhephaestus]

Now the errors will be :

initialise_event_info_struct: eis->send_event == NULL!

initialise_event_info_struct: eis->fd >= FD_SETSIZE!

initialise_event_info_struct: eis->fd <= 0!

[wborn]

For me when the error occurs it results in:

initialise_event_info_struct: eis->fd >= FD_SETSIZE!

[madhephaestus]

This issue seems to be related to this:

https://access.redhat.com/solutions/488623

[madhephaestus]

It seems we would need to transition away from SELECT() and switch to poll()

http://man7.org/linux/man-pages/man2/select.2.html

to

http://man7.org/linux/man-pages/man2/poll.2.html

This will result in lots of opening and closing code needing to be checked when its changed, and a lot of changes to the data structures used to keep track of the serial port events.

[MrDOS]

I agree that switching to poll(2) would be preferable for many reasons, but at the same time, FD_SETSIZE is 1024 just about everywhere. That's a lot of fds, even with 32 physical ports. @roxkamx and @wborn, do your applications churn through opening/closing the serial port many times, or do you typically acquire a port instance once and then hold it open for as long as you can? I'm trying to understand if fd churn is happening due to library or application behaviour. Regardless, this is clearly bad behaviour on the part of the library, but I wonder if we can't find an application-level workaround in the meantime.

[wborn]

It seems the library is keeping thousands of file descriptors to temporary files, e.g. lsof -p PID lists these lines:

java    15241 wouter  955u   REG   0,26         0      947 /run/lock/tmpj1rrF2 (deleted)
java    15241 wouter  956u   REG   0,26         0      948 /run/lock/tmpPwuMWV (deleted)
java    15241 wouter  957u   REG   0,26         0      949 /run/lock/tmpgPk8dP (deleted)
java    15241 wouter  958u   REG   0,26         0      950 /run/lock/tmp356uvI (deleted)
java    15241 wouter  959u   REG   0,26         0      951 /run/lock/tmpKAGSMB (deleted)
java    15241 wouter  960u   REG   0,26         0      952 /run/lock/tmpE64g4u (deleted)
java    15241 wouter  961u   REG   0,26         0      953 /run/lock/tmpxsjGlo (deleted)
java    15241 wouter  962u   REG   0,26         0      954 /run/lock/tmpKDm6Ch (deleted)
java    15241 wouter  963u   REG   0,26         0      955 /run/lock/tmpmLexUa (deleted)
java    15241 wouter  964u   REG   0,26         0      956 /run/lock/tmpuMVYb4 (deleted)
java    15241 wouter  965u   REG   0,26         0      957 /run/lock/tmpSNrrtX (deleted)
java    15241 wouter  966u   REG   0,26         0      958 /run/lock/tmp8FMUKQ (deleted)
java    15241 wouter  967u   REG   0,26         0      959 /run/lock/tmpLbYo2J (deleted)
java    15241 wouter  968u   REG   0,26         0      960 /run/lock/tmpKxYTjD (deleted)
java    15241 wouter  969u   REG   0,26         0      961 /run/lock/tmpGWOpBw (deleted)
java    15241 wouter  970u   REG   0,26         0      962 /run/lock/tmpLTtWSp (deleted)
java    15241 wouter  971u   REG   0,26         0      963 /run/lock/tmpcAWtaj (deleted)
java    15241 wouter  972u   REG   0,26         0      964 /run/lock/tmptSf2rc (deleted)
java    15241 wouter  973u   REG   0,26         0      965 /run/lock/tmps2oBJ5 (deleted)
java    15241 wouter  974u   REG   0,26         0      966 /run/lock/tmp5pfb1Y (deleted)
java    15241 wouter  975u   REG   0,26         0      967 /run/lock/tmp6fWLiS (deleted)
java    15241 wouter  976u   REG   0,26         0      968 /run/lock/tmpGatnAL (deleted)

It's easily reproduced by running the example program in #180 and not inserting the USB stick. If I insert it after there are thousands of these file descriptors I immediately see the error. I think I've even seen it cause issues with other programs because of this.