Kernel panic in dovecot-ee-lmtp on Debian 8

[Tobi]

Hi list

I just realized that that I don not receive all mails in my mailbox
(running dovecot-ee 2.2.23.1-1 on Debian 8). On my frontend servers
(running postfix) the queue fills up with mails that cannot be delivered
via lmtp to my backend servers. The error message on the frontend is
"lost connection with backend while sending data"
When I checked the logs on the backend server I found upon every
delivery attempt kernel panics

Apr 25 12:33:36 mbox1 dovecot: lmtp(REDACTED): Panic: epoll_ctl(del, 18)
failed: Bad file descriptor
Apr 25 12:33:36 mbox1 dovecot: lmtp(REDACTED): Error: Raw backtrace:
/usr/lib/dovecot/libdovecot.so.0(+0x85dee) [0x7faa19225dee] ->
/usr/lib/dovecot/libdovecot.so.0(+0x85edc) [0x7faa19225edc] ->
/usr/lib/dovecot/libdovecot.so.0(i_fatal+0) [0x7faa191ca5be] ->
/usr/lib/dovecot/libdovecot.so.0(+0x9afc1) [0x7faa1923afc1] ->
/usr/lib/dovecot/libdovecot.so.0(+0x994af) [0x7faa192394af] ->
/usr/lib/dovecot/libdovecot.so.0(+0xa5eda) [0x7faa19245eda] ->
/usr/lib/dovecot/libdovecot.so.0(+0xa4566) [0x7faa19244566] ->
/usr/lib/dovecot/libdovecot.so.0(o_stream_destroy+0xe) [0x7faa192446ee]
-> /usr/lib/dovecot/libdovecot-sieve.so.0(+0x7e44b) [0x7faa17d3a44b] ->
/usr/lib/dovecot/libdovecot-sieve.so.0(+0x7ea14) [0x7faa17d3aa14] ->
/usr/lib/dovecot/libdovecot.so.0(+0xa6d88) [0x7faa19246d88] ->
/usr/lib/dovecot/libdovecot.so.0(io_loop_call_io+0x4c) [0x7faa19239c6c]
-> /usr/lib/dovecot/libdovecot.so.0(io_loop_handler_run_internal+0x101)
[0x7faa1923b0d1] ->
/usr/lib/dovecot/libdovecot.so.0(io_loop_handler_run+0x25)
[0x7faa19239cf5] -> /usr/lib/dovecot/libdovecot.so.0(io_loop_run+0x38)
[0x7faa19239e98] ->
/usr/lib/dovecot/libdovecot-sieve.so.0(program_client_run+0xe8)
[0x7faa17d3b308] ->
/usr/lib/dovecot/modules/sieve/lib90_sieve_extprograms_plugin.so(+0x45b8) [0x7faa15d855b8]
->
/usr/lib/dovecot/libdovecot-sieve.so.0(sieve_interpreter_continue+0x7c)
[0x7faa17cfa2ac] ->
/usr/lib/dovecot/libdovecot-sieve.so.0(sieve_interpreter_run+0x2b)
[0x7faa17cfa4cb] -> /usr/lib/dovecot/libdovecot-sieve.so.0(+0x519e0)
[0x7faa17d0d9e0] ->
/usr/lib/dovecot/libdovecot-sieve.so.0(sieve_execute+0x47)
[0x7faa17d0e567] ->
/usr/lib/dovecot/modules/lib90_sieve_plugin.so(+0x3b6b) [0x7faa17f73b6b]
-> /usr/lib/dovecot/libdovecot-lda.so.0(mail_deliver+0x49)
[0x7faa197e5849] -> dovecot/lmtp() [0x4064b3] -> dovecot/lmtp()
[0x406cb4] -> /usr/lib/dovecot/libdovecot.so.0(io_loop_call_io+0x4c)
[0x7faa19239c6c] ->
/usr/lib/dovecot/libdovecot.so.0(io_loop_handler_run_internal+0x101)
[0x7faa1923b0d1] ->
/usr/lib/dovecot/libdovecot.so.0(io_loop_handler_run+0x25) [0x7faa19239cf5]
Apr 25 12:33:36 mbox1 dovecot: lmtp(REDACTED): Fatal: master:
service(lmtp): child 32081 killed with signal 6 (core dumps disabled)

the weird thing is that this is not concerning all incoming mails. From
time to time the frontends can deliver messages to my backends. I just
tested it with a mail from work. It arrived via a frontend that has
other mails in queue that could not be delivered so far.

Does anyone have an idea where I could look for the root cause of this
panic messages?

Thanks a lot

tobi