Cellular Network Infrastructure » Radio Access Network » OsmoBSC

hmm, the conclusion seems right, but let's first try to find out which conn it is that has no lchan, and whether that makes sense at all.
The fact that this testcase alone cannot trigger the failure indicates that some previous test leaves a conn lingering behind.

On branch neels/hack I pushed a patch that might help to find out:

commit e92ce3f0c6a03084c9f1037204c94c9a44c37bdd (origin/neels/hack, neels/hack)
Author: Neels Hofmeyr <neels@hofmeyr.de>
Date:   Tue Nov 27 18:19:33 2018 +0100

    NOT FOR MERGE: try to find out why there is a conn without an lchan

    in ttcn3 tests, we currently face a crash where the vty 'handover any' command
    assumes all conns have an lchan.

    Don't access a NULL lchan.

    But first we want to find out why there is a NULL lchan, so log a visible error.

    Change-Id: If5909b3890cf3f6327b271e53e458a417cebf73f

diff --git a/src/osmo-bsc/bsc_vty.c b/src/osmo-bsc/bsc_vty.c
index 983dcb9b3..947ba67df 100644
--- a/src/osmo-bsc/bsc_vty.c
+++ b/src/osmo-bsc/bsc_vty.c
@@ -1665,6 +1665,11 @@ static int ho_or_as(struct vty *vty, const char *argv[], int argc)

        /* Find the connection/lchan that we want to handover */
        llist_for_each_entry(conn, &net->subscr_conns, entry) {
+               if (!conn->lchan) {
+                       LOGP(DHO, LOGL_ERROR, "XXXXX found a conn without an lchan: %s %s\n",
+                            osmo_fsm_inst_name(conn->fi), bsc_subscr_name(conn->bsub));
+                       continue;
+               }
                if (conn_get_bts(conn)->nr == bts_nr &&
                    conn->lchan->ts->trx->nr == trx_nr &&
                    conn->lchan->ts->nr == ts_nr && conn->lchan->nr == ss_nr) {

daniel can you try to reproduce the problem with this patch applied? Then we might see an error log with XXXXX telling us the FSM instance name of a conn, and hence we might be able to associate that with some earlier logging from osmo-bsc to find out why that conn is still around.

With your patch the only failing test is currently TC_paging_resp_unsol:
f_expect_chan_rel(): expect_rll_rel_req=true got_rll_rel_req=false

I can see the log message you added:

Wed Nov 28 14:37:47 2018 DHO <0008> bsc_vty.c:1670 XXXXX found a conn without an lchan: SUBSCR_CONN[0x21d7010] unknown

Lines 32073 of osmo-bsc.log show where this came from:

32073 Wed Nov 28 14:36:11 2018 DRSL <0003> osmo_bsc_main.c:283 bootstrapping RSL for BTS/TRX (0/0) on ARFCN 871 using MCC-MNC 001-01 LAC=1 CID=0  32073  BSIC=10
32074 Wed Nov 28 14:36:12 2018 DRSL <0003> abis_rsl.c:1357 (bts=0) CHAN RQD: reason: answer to paging (ra=0x23, neci=0x01, chreq_reason=0x01)
32075 Wed Nov 28 14:36:12 2018 DMSC <0007> fsm.c:299 SUBSCR_CONN[0x21d7010]{INIT}: Allocated
32076 Wed Nov 28 14:36:12 2018 DLCLS <000f> fsm.c:299 LCLS[0x21d7140]{NO_LCLS}: Allocated
32077 Wed Nov 28 14:36:12 2018 DLCLS <000f> fsm.c:329 LCLS[0x21d7140]{NO_LCLS}: is child of SUBSCR_CONN[0x21d7010]
32078 Wed Nov 28 14:36:12 2018 DMSC <0007> gsm_08_08.c:328 Got paged but no subscriber found.
32079 Wed Nov 28 14:36:12 2018 DMSC <0007> gsm_08_08.c:363 Failed to find a MSC for a connection.
32080 Wed Nov 28 14:36:12 2018 DMSC <0007> lchan_fsm.c:1390 SUBSCR_CONN[0x21d7010]{INIT}: lchan lchan(0-0-0-CCCH_SDCCH4-0)[0x21d8a10] detaches from conn

The failing test TC_paging_resp_unsol seems to be responsible for the that.

It seems I have forgotten about this issue; all I know is that currently this test passes for osmo-bsc master