Hello Daryl,
I can reproduce locally the slow DS startup (due to slapi-nis
priming). In fact the version I was using had not the slapi-nis fix
to differ the priming.
I failed to reproduce the intensive load on DS when krb5kdc startup.
Looking at yours logs, we can see that krb5kdc startup triggers a
set of requests during 3s up to 8s. The logs are looking like (note
the etime can go up to 2s):
[10/Mar/2016:14:20:35 -0600] conn=40 fd=87 slot=87 connection from
local to /var/run/slapd-UOFMT1.socket
[10/Mar/2016:14:20:36 -0600] conn=40 AUTOBIND dn="cn=Directory Manager"
[10/Mar/2016:14:20:36 -0600] conn=40 op=0 BIND dn="cn=Directory
Manager" method=sasl version=3 mech=EXTERNAL
[10/Mar/2016:14:20:36 -0600] conn=40 op=0 RESULT err=0 tag=97
nentries=0 *etime=1* dn="cn=Directory Manager"
[10/Mar/2016:14:20:36 -0600] conn=40 op=1 SRCH
base="cn=UOFMT1,cn=kerberos,dc=uofmt1" scope=0
filter="(objectClass=*)" attrs=ALL
[10/Mar/2016:14:20:36 -0600] conn=40 op=1 RESULT err=0 tag=101
nentries=1 etime=0
[10/Mar/2016:14:20:36 -0600] conn=40 op=2 SRCH
base="cn=ipaConfig,cn=etc,dc=uofmt1" scope=0
filter="(objectClass=*)" attrs="ipaConfigString ipaKrbAuthzData
ipaUserAuthType"
[10/Mar/2016:14:20:36 -0600] conn=40 op=2 RESULT err=0 tag=101
nentries=1 etime=0
[10/Mar/2016:14:20:36 -0600] conn=40 op=3 SRCH base="dc=uofmt1"
scope=2 filter="(objectClass=ipaNTDomainAttrs)" attrs="ipaNTFlatName
ipaNTFallbackPrimaryGroup ipaNTSecurityIdentifier"
[10/Mar/2016:14:20:37 -0600] conn=40 op=3 RESULT err=0 tag=101
nentries=0 *etime=1*
[10/Mar/2016:14:20:37 -0600] conn=40 op=4 SRCH
base="cn=UOFMT1,cn=kerberos,dc=uofmt1" scope=0 filter="(krbMKey=*)"
attrs="krbMKey"
[10/Mar/2016:14:20:37 -0600] conn=40 op=4 RESULT err=0 tag=101
nentries=1 etime=0
[10/Mar/2016:14:20:37 -0600] conn=40 op=5 SRCH base="dc=uofmt1"
scope=2
filter="(&(|(objectClass=krbprincipalaux)(objectClass=krbprincipal))(krbPrincipalName=K/M@UOFMT1))"
attrs="krbPrincipalName krbCanonicalName ipaKrbPrincipalAlias
krbUPEnabled krbPrincipalKey krbTicketPolicyReference
krbPrincipalExpiration krbPasswordExpiration krbPwdPolicyReference
krbPrincipalType krbPwdHistory krbLastPwdChange krbPrincipalAliases
krbLastSuccessfulAuth krbLastFailedAuth krbLoginFailedCount
krbExtraData krbLastAdminUnlock krbObjectReferences krbTicketFlags
krbMaxTicketLife krbMaxRenewableAge nsAccountLock passwordHistory
ipaKrbAuthzData ipaUserAuthType ipatokenRadiusConfigLink objectClass"
[10/Mar/2016:14:20:38 -0600] conn=40 op=5 RESULT err=0 tag=101
nentries=1 *etime=1*
[10/Mar/2016:14:20:39 -0600] conn=40 op=6 UNBIND
[10/Mar/2016:14:20:39 -0600] conn=40 op=6 fd=87 closed - U1
I think the request op=3 (SRCH base="dc=uofmt1" scope=2
filter="(objectClass=ipaNTDomainAttrs)") is slow also because of
slapi-nis. In fact it is indexed and returns 0 entry. So only
plugins can create this high etime.
An improvement in slapi-nis makes its search callback noop when it
comes from krb and I am running this improvement.
In conclusion I think both slow DS startup and KRB5 startup are
fixed in RHEL 7.
thanks
theirry
On 03/14/2016 08:46 PM, Daryl Fonseca-Holt wrote:
Hello Thierry,
Attached is the pstacks from only the final DS restart. I don't think
they will show the whole picture.
According to the debug log /var/log/ipareplica-install.log (attached)
the start of the krb5kdc.service (19:13:16Z) is successful, but the
krb5kdc log (attach) shows it is unable to fetch the master K/M at
14:31:31CDT (-5hour offset). This is when the install log shows
kadmind failing.
In my experience with the master observing top there are two intense
times for ns-slapd-<instance>. The first when it start, of course, and
the second when krb5kdc starts. I assume this is because krb5kdc must
get it's configuration and data from the same DS. krb5kdc fails but
the ipareplica-install script isn't aware of it. Finally
kadmin.service tries to access krb5kdc and finds that it is dead.
Please note these logs are with Schema Compatability and NIS plugins
turned off per the other e-mail from Alexander.
I've noticed on a running master I can prevent this type of failure by
manually starting dirsrv (systemctl start dirsrv@<instance>.service),
watch top until all threads of ns-slapd have settled, then systemctl
start krb5kdc.service, again watching top until ns-slapd threads have
settled down before systemctl start kadmin.service. This kind of
manual intervention is is not possible when running the
ipareplica-install script.
I will look into introducing a delay at the completion of the dirsrv
and krb5kdc systemd units and see if I can accommodate
ipareplica-install. Just as an experiment for now. I need to advance
the project into High Availability testing but cannot do so without a
functioning replica.
Regards, Daryl
On 03/14/16 09:20, thierry bordaz wrote:
Hi Daryl,
Thanks for all the data. I will look at the pstacks. A first look
shows that you capture import, bind... so may be a complete
ipa-replica-install session.
I will try to retrieve the specific startup time to see what was
going on at that time.
If you have the time to monitor only startup, it will help me
shrinking the set of pstacks.
Startup of DS last > 1min. If you may start DS and as soon as the
ns-slapd process is launched, do regular pstacks. Then when you are
able to send a simple ldapsearch (ldapsearch -x -b "" -s base), you
may stop taking pstacks.
thanks
thierry
On 03/14/2016 03:06 PM, Daryl Fonseca-Holt wrote:
Hi Thierry,
I moved the old logs into a subdirectory called try1. I did the
recommended ipa-server-install --uninstall. Tried the replica
install again. Failed during kadmind start like the previous time.
The log from ipa-replica-install (with -d) is at
http://home.cc.umanitoba.ca/~fonsecah/ipa/ipareplica-install.log
The console script (mostly the same as the log but with my entries)
is at
http://home.cc.umanitoba.ca/~fonsecah/ipa/ipa-replica-install.console
The 5 second pstacks are at
http://home.cc.umanitoba.ca/~fonsecah/ipa/slapd-pstacks.console
Thanks, Daryl
On 03/11/16 02:40, thierry bordaz wrote:
Hello Deryl,
My understanding is that ns-slapd is first slow to startup.
Then when krb5kdc is starting it may load ns-slapd.
We identified krb5kdc may be impacted by the number of users
accounts.
From the ns-slapd errors log it is not clear why it is so slow
to start.
Would you provide the ns-slapd access logs from that period.
Also in order to know where ns-slapd is spending time, it would
really help if you can get regular (each 5s) pstacks (with
389-ds-debuginfo), during DS startup and then later during
krb5kdc startup.
best regards
thierry
On 03/10/2016 11:10 PM, Daryl Fonseca-Holt wrote:
Environment:
RHEL 7.2
IPA 4.2.0-15
nss 3.19.1-19
389-ds-base 1.3.4.0-26
sssd 1.13.0-40
I've encountered this problem in IPA 3.0.0 but hoped it was
addressed in 4.2.0.
Trying to set up a replica of a master with 150,000+ user
accounts, NIS and Schema Compatability enabled on the master.
During ipa-replica-install it attempts to start IPA. dirsrv
starts, krb5kdc starts, but then kadmind fails because krb5kdc has
gone missing.
This happens during restart of IPA in version 3.0.0 too. There it
can be overcome by manually starting each component of IPA _but_
waiting until ns-slapd-<instance> has settled down (as seen from
top) before starting krb5kdc. I also think that the startup of
krb5kdc loads the LDAP instance quite a bit.
There is a problem in the startup logic where dirsrv is so busy
that even though krb5kdc successfully starts and allows the kadmin
to begin kdb5kdc is not really able to do its duties.
I'm reporting this since there must be some way to delay the start
of krb5kdc and then kadmind until ns-slapd-<instance> is really
open for business.
# systemctl status krb5kdc.service
● krb5kdc.service - Kerberos 5 KDC
Loaded: loaded (/usr/lib/systemd/system/krb5kdc.service;
disabled; vendor preset: disabled)
Active: inactive (dead)
Mar 10 14:19:13 jutta.cc.umanitoba.ca systemd[1]: Stopped Kerberos
5 KDC.
Mar 10 14:20:36 jutta.cc.umanitoba.ca systemd[1]: Starting
Kerberos 5 KDC...
Mar 10 14:20:39 jutta.cc.umanitoba.ca systemd[1]: Started Kerberos
5 KDC.
# systemctl status krb5kdc.service
● krb5kdc.service - Kerberos 5 KDC
Loaded: loaded (/usr/lib/systemd/system/krb5kdc.service;
disabled; vendor preset: disabled)
Active: inactive (dead)
Mar 10 14:19:13 jutta.cc.umanitoba.ca systemd[1]: Stopped Kerberos
5 KDC.
Mar 10 14:20:36 jutta.cc.umanitoba.ca systemd[1]: Starting
Kerberos 5 KDC...
Mar 10 14:20:39 jutta.cc.umanitoba.ca systemd[1]: Started Kerberos
5 KDC.
journalctl -xe was stale by the time I got to it so I've attached
/var/log/messages instead.
The log from ipa-replica-install (with -d) is at
http://home.cc.umanitoba.ca/~fonsecah/ipa/ipareplica-install.log
The console script (mostly the same as the log but with my
entries) is at
http://home.cc.umanitoba.ca/~fonsecah/ipa/ipa-replica-install.console
The /var/log/dirsrv/ns-slapd-<instance> access log is at
http://home.cc.umanitoba.ca/~fonsecah/ipa/access
Regards, Daryl
--
--
Daryl Fonseca-Holt
IST/CNS/Unix Server Team
University of Manitoba
204.480.1079
--
--
Daryl Fonseca-Holt
IST/CNS/Unix Server Team
University of Manitoba
204.480.1079
--
Manage your subscription for the Freeipa-users mailing list:
https://www.redhat.com/mailman/listinfo/freeipa-users
Go to http://freeipa.org for more info on the project
