[Freeipa-users] replica DS failure deadlock
thierry bordaz
tbordaz at redhat.com
Wed Oct 19 17:05:14 UTC 2016
On 10/19/2016 06:54 PM, Andrew E. Bruno wrote:
> On Wed, Oct 19, 2016 at 06:33:05PM +0200, thierry bordaz wrote:
>>
>> On 10/19/2016 03:48 PM, Andrew E. Bruno wrote:
>>> On Wed, Oct 19, 2016 at 10:13:26AM +0200, Ludwig Krispenz wrote:
>>>> On 10/18/2016 08:52 PM, Andrew E. Bruno wrote:
>>>>> We had one of our replicas fail today with the following errors:
>>>>>
>>>>>
>>>>> [18/Oct/2016:13:40:47 -0400] agmt="cn=meTosrv-m14-32.cbls.ccr.buffalo.edu" (srv-m14-32:389) - Can't locate CSN 58065ef3000100030000 in the changelog (DB rc=-30988). If replication stops, the consumer may need to be reinitialized.
>>>>> [18/Oct/2016:13:43:07 -0400] NSMMReplicationPlugin - changelog program - _cl5WriteOperationTxn: retry (49) the transaction (csn=58065f74000500040000) failed (rc=-30993 (BDB0068 DB_LOCK_DEADLOCK: Locker killed to resolve a deadlock))
>>>>> [18/Oct/2016:13:43:07 -0400] NSMMReplicationPlugin - changelog program - _cl5WriteOperationTxn: failed to write entry with csn (58065f74000500040000); db error - -30993 BDB0068 DB_LOCK_DEADLOCK: Locker killed to resolve a deadlock
>>>>> [18/Oct/2016:13:43:07 -0400] NSMMReplicationPlugin - write_changelog_and_ruv: can't add a change for uid=janedoe,cn=users,cn=accounts,dc=cbls,dc=ccr,dc=buffalo,dc=edu (uniqid: 939bca48-2ced11e5-ac0b8f7e-e0b1a377, optype: 64) to changelog csn 58065f74000500040000
>>>>> [18/Oct/2016:13:43:07 -0400] - SLAPI_PLUGIN_BE_TXN_POST_MODRDN_FN plugin returned error but did not set SLAPI_RESULT_CODE
>>>>> [18/Oct/2016:13:43:07 -0400] NSMMReplicationPlugin - process_postop: Failed to apply update (58065f74000500040000) error (1). Aborting replication session(conn=1314106 op=1688559)
>>>>> [18/Oct/2016:13:43:12 -0400] - cos_cache_change_notify: modified entry is NULL--updating cache just in case
>>>>> [18/Oct/2016:13:43:12 -0400] - Skipping CoS Definition cn=Password Policy,cn=accounts,dc=cbls,dc=ccr,dc=buffalo,dc=edu--no CoS Templates found, which should be added before the CoS Definition.
>>>>> [18/Oct/2016:13:43:20 -0400] - Operation error fetching Null DN (4a729f9a-955a11e6-aaffa516-e778e883), error -30993.
>>>>> [18/Oct/2016:13:43:20 -0400] - dn2entry_ext: Failed to get id for changenumber=30856302,cn=changelog from entryrdn index (-30993)
>>>>> [18/Oct/2016:13:43:20 -0400] - Operation error fetching changenumber=30856302,cn=changelog (null), error -30993.
>>>>> [18/Oct/2016:13:43:20 -0400] DSRetroclPlugin - replog: an error occured while adding change number 30856302, dn = changenumber=30856302,cn=changelog: Operations error.
>>>>> [18/Oct/2016:13:43:20 -0400] retrocl-plugin - retrocl_postob: operation failure [1]
>>>>> [18/Oct/2016:13:43:20 -0400] NSMMReplicationPlugin - process_postop: Failed to apply update (58065f9f000000600000) error (1). Aborting replication session(conn=1901274 op=5)
>>>>> [18/Oct/2016:13:43:24 -0400] - ldbm_back_seq deadlock retry BAD 1601, err=0 BDB0062 Successful return: 0
>>>>> [18/Oct/2016:13:43:25 -0400] NSMMReplicationPlugin - changelog program - _cl5WriteOperationTxn: retry (49) the transaction (csn=58065f7c000a00040000) failed (rc=-30993 (BDB0068 DB_LOCK_DEADLOCK: Locker killed to resolve a deadlock))
>>>>> [18/Oct/2016:13:43:25 -0400] NSMMReplicationPlugin - changelog program - _cl5WriteOperationTxn: failed to write entry with csn (58065f7c000a00040000); db error - -30993 BDB0068 DB_LOCK_DEADLOCK: Locker killed to resolve a deadlock
>>>>> [18/Oct/2016:13:43:25 -0400] NSMMReplicationPlugin - write_changelog_and_ruv: can't add a change for uid=janedoe,cn=users,cn=accounts,dc=cbls,dc=ccr,dc=buffalo,dc=edu (uniqid: 4080421a-2d0211e5-ac0b8f7e-e0b1a377, optype: 64) to changelog csn 58065f7c000a00040000
>>>>>
>>>>>
>>>>> ns-slapd was hung so we restarted and now it's stuck and won't come back up. It
>>>>> hangs up here:
>>>>>
>>>>> [18/Oct/2016:14:12:31 -0400] - Skipping CoS Definition cn=Password Policy,cn=accounts,dc=cbls,dc=ccr,dc=buffalo,dc=edu--no CoS Templates found, which should be added before the CoS Definition.
>>>>> [18/Oct/2016:14:12:31 -0400] NSMMReplicationPlugin - changelog program - _cl5NewDBFile: PR_DeleteSemaphore: /var/lib/dirsrv/slapd-CBLS-CCR-BUFFALO-EDU/cldb/a32992ce-71b811e5-9d33a516-e778e883.sema; NSPR error - -5943
>>>>> [18/Oct/2016:14:12:32 -0400] NSMMReplicationPlugin - changelog program - _cl5NewDBFile: PR_DeleteSemaphore: /var/lib/dirsrv/slapd-CBLS-CCR-BUFFALO-EDU/cldb/986efe12-71b811e5-9d33a516-e778e883.sema; NSPR error - -5943
>>>>>
>>>>>
>>>>> Tried deleting the semaphore files and restarting but no luck. Attached
>>>>> is a stacktrace of the stuck ns-slapd process.
>>>>>
>>>>> Here's the versions were running:
>>>>>
>>>>> ipa-server-4.2.0-15.0.1.el7.centos.19.x86_64
>>>>> 389-ds-base-1.3.4.0-33.el7_2.x86_64
>>>>>
>>>>> FWIW, we were experimenting with the new life-cycle management features,
>>>>> specifically "preserved" users and deleted the user "janedoe" when this
>>>>> happened. From the errors above looks like this host failed to
>>>>> replicate the change? Not sure if this is related or not.
>>>>>
>>>>> Is it possible to recover the database? Thanks in advance for any pointers.
>>>> from the stack trace the process is not hanging, it is trying to recover.
>>>> After a crash/kill the changelog does not contai a RUV and it is
>>>> reconstructed by reading all records in the changelog, if this is large it
>>>> can take some time.
>>>> If you look at that part of the stack repeatedly,
>>>>
>>>> #4 0x00007f4e88daeba5 in cl5DBData2Entry (data=<optimized out>, len=<optimized out>, entry=entry at entry=0x7ffff6598910) at ldap/servers/plugins/replication/cl5_api.c:2342
>>>> rc = <optimized out>
>>>> version = <optimized out>
>>>> pos = 0x7f4e9839d091 ""
>>>> strCSN = 0x0
>>>> op = 0x7ffff6598980
>>>> add_mods = 0x7f4e983a5e80
>>>> rawDN = 0x7f4e98396e20 "fqdn=cpn-k08-29-02.cbls.ccr.buffalo.edu,cn=computers,cn=accounts,dc=cbls,dc=ccr,dc=buffalo,dc=edu"
>>>> s = "\300\037>\230N\177\000\000@\210Y\366\377\177\000\000@\210Y\366\377"
>>>> #5 0x00007f4e88daf5d6 in _cl5GetNextEntry (entry=entry at entry=0x7ffff6598910, iterator=0x7f4e983a5e80) at ldap/servers/plugins/replication/cl5_api.c:5291
>>>> rc = 0
>>>> it = 0x7f4e983a5e80
>>>> key = {data = 0x0, size = 21, ulen = 0, dlen = 0, doff = 0, app_data = 0x0, flags = 16}
>>>> data = {data = 0x7f4e9839cff0, size = 335, ulen = 0, dlen = 0, doff = 0, app_data = 0x0, flags = 16}
>>>> #6 0x00007f4e88dafb34 in _cl5ConstructRUV (purge=1, obj=0x7f4e983e1fc0, replGen=0x7ffff6598910 "\200\211Y\366\377\177") at ldap/servers/plugins/replication/cl5_api.c:4306
>>>>
>>>>
>>>> you should see some progress in which entry is handled
>>>>
>>> Ludwig, thanks very much for the help. As you pointed out just needed to let it
>>> finish. ns-slapd eventually came back up once it finished reading the
>>> changelog. Still seeing some errors related to the NSMMReplicationPlugin failed
>>> to apply update and from the managed-entries-plugin. Can these safely be
>>> ignored or are they indicative of a more serious problem?
>> This is difficult to say the reason of managed entries messages.
>> It says that the origin entry "uid=janedoe,cn=deleted
>> users,cn=accounts,cn=provisioning,dc=cbls,dc=ccr,dc=buffalo,dc=edu"
>> is still having a managed entry ('|mepManagedEntry') that is possibly
>> something like
>> '|cn=janedoe,cn=groups,cn=accounts,dc=cbls,dc=ccr,dc=buffalo,dc=edu".
>>
>> This is looking like a bug because user 'janedoe' being a preserved user, it
>> should not have any reference to existing groups.
>>
>> Could you dump uid=janedoe entry:
>> ldapsearch -D "cn=directory manager" -w xxxx -b ""uid=janedoe,cn=deleted
>> users,cn=accounts,cn=provisioning,dc=cbls,dc=ccr,dc=buffalo,dc=edu"
>> nscpentrywsi
> Here's the entry for janedoe:
>
> ldapsearch -Y GSSAPI -b "uid=janedoe,cn=deleted users,cn=accounts,cn=provisioning,dc=cbls,dc=ccr,dc=buffalo,dc=edu" nscpentrywsi
> # extended LDIF
> #
> # LDAPv3
> # base <uid=janedoe,cn=deleted users,cn=accounts,cn=provisioning,dc=cbls,dc=ccr,dc=buffalo,dc=edu> with scope subtree
> # filter: (objectclass=*)
> # requesting: nscpentrywsi
> #
>
> # janedoe, deleted users, accounts, provisioning, cbls.ccr.buffalo.edu
> dn: uid=janedoe,cn=deleted users,cn=accounts,cn=provisioning,dc=cbls,dc=ccr,d
> c=buffalo,dc=edu
>
> # search result
> search: 4
> result: 0 Success
>
> # numResponses: 2
> # numEntries: 1
nscpentrywsi is a specific attribute that dumps the entry. It is only
available for 'cn=directory manager' but not for 'admin'.
If you do not know the 'cn=directory manager' password, then being
'admin' do the same request without specifying any attributes
ldapsearch -Y GSSAPI -LLL -s base -b "uid=janedoe,cn=deleted users,cn=accounts,cn=provisioning,dc=cbls,dc=ccr,dc=buffalo,dc=edu"
>
>
>> If the link still exists, it is looking like a bug but IMHO it should not
>> create security issue.
>>
>> regards
>> thierry
More information about the Freeipa-users
mailing list