Ceph » rgw

While trying to deploy a rgw node with ceph-ansible using keystone authentication I noticed that the radosgw process repeatedly crashes. While investigating I narrowed it down to the 'nss db path' setting in ceph.conf.

radosgw starts up fine with following ceph.conf:

[global]
cluster network = 10.129.xxx.0/24
fsid = c0f9f405-c55a-494c-b03c-04150b6d25c1
mon host = 10.129.xxx.xx,10.129.xxx.xx,10.129.xxx.xx,10.129.xxx.xx
osd_pool_default_pg_num = 16
public network = 10.129.xxx.0/23
rgw bucket default quota max objects = 1638400
rgw override bucket index max shards = 16

[client.rgw.radosgw-01]
host = radosgw-01
keyring = /var/lib/ceph/radosgw/ceph-rgw.radosgw-01/keyring
log file = /var/log/ceph/ceph-rgw-radoswg-01.log
rgw frontends = beast endpoint=10.129.xxx.xxx:8080

When adding the 'nss db path' it crashes:
[global]
cluster network = 10.129.xxx.0/24
fsid = c0f9f405-c55a-494c-b03c-04150b6d25c1
mon host = 10.129.xxx.xx,10.129.xxx.xx,10.129.xxx.xx,10.129.xxx.xx
osd_pool_default_pg_num = 16
public network = 10.129.xxx.0/23
rgw bucket default quota max objects = 1638400
rgw override bucket index max shards = 16

[client.rgw.radosgw-01]
host = radosgw-01
keyring = /var/lib/ceph/radosgw/ceph-rgw.radosgw-01/keyring
log file = /var/log/ceph/ceph-rgw-radoswg-01.log
rgw frontends = beast endpoint=10.129.xxx.xxx:8080
nss db path = /var/lib/ceph/radosgw/ceph-rgw.radosgw-01/nss

Stacktrace:
ceph version 13.2.2 (02899bfda814146b021136e9d8e80eba494e1126) mimic (stable)
#1 0x00007fd191444801 in _GI_abort () at abort.c:79
#2 0x00007fd1930bb710 in ceph::_ceph_assert_fail(char const*, char const*, int, char const*) () from /usr/lib/ceph/libceph-common.so.0
#3 0x00007fd1930bb787 in ceph::__ceph_assert_fail(ceph::assert_data const&) () from /usr/lib/ceph/libceph-common.so.0
#4 0x00007fd1933ab5ab in ceph::crypto::init(CephContext*) () from /usr/lib/ceph/libceph-common.so.0
#5 0x00007fd193110fd3 in MonClient::get_monmap_and_config() () from /usr/lib/ceph/libceph-common.so.0
#6 0x000055843a451812 in global_pre_init (defaults=<optimized out>, args=std::vector of length 0, capacity 7, module_type=<optimized out>, code_env=<optimized out>, flags=1)
at ./src/global/global_init.cc:150
#7 0x000055843a3a548d in main (argc=8, argv=0x7ffee96369b8) at ./src/rgw/rgw_main.cc:197

I tested on a recent Ubuntu 18.04 and CentOS 7, it didn't matter if the nss db path directory existed or not, 'nss-tools/libnss3-tools' were present on the machines.