Re: [Freeipa-devel] [PATCH] 1068 wait for LDAP when renewing the RA
Hi, On 24.10.2012 21:24, Rob Crittenden wrote: All the certs are pretty critical in certificate renewal but the agent cert has the distinction of having to be updated in multiple places. It needs to exist in both LDAP servers. It is possible that one or both of these servers may be down briefly during renewal so we need to be a bit more robust in our handling. This will wait up to 5 minutes per server to try to update things, and syslog when failures occur. It is now also safe to re-run this in case something catastrophic happens. One would just need to manually run this to load the required data into LDAP. rob I believe that there should be a break statement after the updated = True line: +updated = True +except errors.NetworkError: It would be nice if this message said 30 s instead of just 30: +syslog.syslog(syslog.LOG_ERR, 'Connection to %s failed, sleeping 30' % dogtag_uri) The continue statement is redundant: +attempts += 1 +continue except errors.EmptyModlist: The variables you access in both of the finally blocks (conn and tmpdir) may be unbound. This can be fixed like this: while attempts 10: conn = None tmpdir = None try: ... finally: if conn is not None and conn.isconnected(): conn.disconnect() if tmpdir is not None: shutil.rmtree(tmpdir) It would be nice if this message (both instances of it) included sys.argv[0], so that it is obvious to the user what script is this script: +syslog.syslog(syslog.LOG_ERR, 'Giving up. This script may be safely re-executed.') No spaces in kwarg assignment please: +tmpdir = tempfile.mkdtemp(prefix = tmp-) You might want to include sleeping 30 s in this message as well: +except Exception, e: +syslog.syslog(syslog.LOG_ERR, 'Updating renewal certificate failed: %s' % e) +time.sleep(30) Honza -- Jan Cholasta ___ Freeipa-devel mailing list Freeipa-devel@redhat.com https://www.redhat.com/mailman/listinfo/freeipa-devel
Re: [Freeipa-devel] [PATCH] 1068 wait for LDAP when renewing the RA
Jan Cholasta wrote:
Hi,
On 24.10.2012 21:24, Rob Crittenden wrote:
All the certs are pretty critical in certificate renewal but the agent
cert has the distinction of having to be updated in multiple places. It
needs to exist in both LDAP servers.
It is possible that one or both of these servers may be down briefly
during renewal so we need to be a bit more robust in our handling. This
will wait up to 5 minutes per server to try to update things, and syslog
when failures occur.
It is now also safe to re-run this in case something catastrophic
happens. One would just need to manually run this to load the required
data into LDAP.
rob
I believe that there should be a break statement after the updated =
True line:
+updated = True
+except errors.NetworkError:
Sure is, added.
It would be nice if this message said 30 s instead of just 30:
+syslog.syslog(syslog.LOG_ERR, 'Connection to %s failed,
sleeping 30' % dogtag_uri)
Sure.
The continue statement is redundant:
+attempts += 1
+continue
except errors.EmptyModlist:
Yup. I used to have code executed after the try/except/finally. Removed.
The variables you access in both of the finally blocks (conn and tmpdir)
may be unbound. This can be fixed like this:
while attempts 10:
conn = None
tmpdir = None
try:
...
finally:
if conn is not None and conn.isconnected():
conn.disconnect()
if tmpdir is not None:
shutil.rmtree(tmpdir)
Good catch, added.
It would be nice if this message (both instances of it) included
sys.argv[0], so that it is obvious to the user what script is this
script:
+syslog.syslog(syslog.LOG_ERR, 'Giving up. This script may be safely
re-executed.')
It is included by syslog:
Nov 1 11:13:24 edsel renew_ra_cert: Connection to ldap://localhost:7389
failed, sleeping 30
No spaces in kwarg assignment please:
+tmpdir = tempfile.mkdtemp(prefix = tmp-)
OK.
You might want to include sleeping 30 s in this message as well:
+except Exception, e:
+syslog.syslog(syslog.LOG_ERR, 'Updating renewal certificate
failed: %s' % e)
+time.sleep(30)
Sure, added that.
I also added a missing update. I added handling for ldap.SERVER_DOWN as
a NetworkError instead of a DatabaseError.
rob
From 8e31742686f5b261e6f6a82f9d49882a0a2750a6 Mon Sep 17 00:00:00 2001
From: Rob Crittenden rcrit...@redhat.com
Date: Tue, 23 Oct 2012 14:07:13 -0400
Subject: [PATCH] Wait for the directory server to come up when updating the
agent certificate.
It is possible that either or both of the LDAP instances are being restarted
during the renewal process. Make the script retry if this is the case.
It is also safe to re-run this script if it fails. It will take the current
ipaCert certificate and attempt to update the agent information in LDAP.
https://fedorahosted.org/freeipa/ticket/3179
---
install/restart_scripts/renew_ra_cert | 100 +++---
ipaserver/plugins/ldap2.py| 3 +
2 files changed, 72 insertions(+), 31 deletions(-)
diff --git a/install/restart_scripts/renew_ra_cert b/install/restart_scripts/renew_ra_cert
index 14cbc114ca9d38697cdb2a24b41ccf2ac2c66389..9d4a5e9c431c5445d4bc361630303015c83e53db 100644
--- a/install/restart_scripts/renew_ra_cert
+++ b/install/restart_scripts/renew_ra_cert
@@ -23,6 +23,7 @@ import sys
import shutil
import tempfile
import syslog
+import time
from ipapython import services as ipaservices
from ipapython.certmonger import get_pin
from ipapython import ipautil
@@ -33,6 +34,7 @@ from ipapython.dn import DN
from ipalib import x509
from ipalib import errors
from ipaserver.plugins.ldap2 import ldap2
+import ldap as _ldap
api.bootstrap(context='restart')
api.finalize()
@@ -53,41 +55,77 @@ except IOError, e:
syslog.syslog(syslog.LOG_ERR, 'Unable to determine PIN for CA instance: %s' % e)
sys.exit(1)
-try:
-conn = ldap2(shared_instance=False, ldap_uri='ldap://localhost:%d' % DEFAULT_DSPORT)
-conn.connect(bind_dn=DN(('cn', 'directory manager')), bind_pw=dm_password)
-(entry_dn, entry_attrs) = conn.get_entry(dn, ['usercertificate'], normalize=False)
-entry_attrs['usercertificate'].append(cert)
-entry_attrs['description'] = '2;%d;%s;%s' % (serial_number, issuer, subject)
-conn.update_entry(dn, entry_attrs, normalize=False)
-conn.disconnect()
-except Exception, e:
-syslog.syslog(syslog.LOG_ERR, 'Updating agent entry failed: %s' % e)
-sys.exit(1)
+attempts = 0
+dogtag_uri='ldap://localhost:%d' % DEFAULT_DSPORT
+updated = False
-# Store it in the IPA LDAP server
-tmpdir = tempfile.mkdtemp(prefix = tmp-)
-try:
-dn = DN(('cn','ipaCert'), ('cn', 'ca_renewal'), ('cn', 'ipa'), ('cn', 'etc'), api.env.basedn)
-principal = str('host/%s@%s' % (api.env.host, api.env.realm))
-ccache = ipautil.kinit_hostprincipal('/etc/krb5.keytab', tmpdir, principal)
-conn =
Re: [Freeipa-devel] [PATCH] 1068 wait for LDAP when renewing the RA
On 1.11.2012 16:32, Rob Crittenden wrote: Jan Cholasta wrote: Hi, On 24.10.2012 21:24, Rob Crittenden wrote: All the certs are pretty critical in certificate renewal but the agent cert has the distinction of having to be updated in multiple places. It needs to exist in both LDAP servers. It is possible that one or both of these servers may be down briefly during renewal so we need to be a bit more robust in our handling. This will wait up to 5 minutes per server to try to update things, and syslog when failures occur. It is now also safe to re-run this in case something catastrophic happens. One would just need to manually run this to load the required data into LDAP. rob I believe that there should be a break statement after the updated = True line: +updated = True +except errors.NetworkError: Sure is, added. It would be nice if this message said 30 s instead of just 30: +syslog.syslog(syslog.LOG_ERR, 'Connection to %s failed, sleeping 30' % dogtag_uri) Sure. The continue statement is redundant: +attempts += 1 +continue except errors.EmptyModlist: Yup. I used to have code executed after the try/except/finally. Removed. The variables you access in both of the finally blocks (conn and tmpdir) may be unbound. This can be fixed like this: while attempts 10: conn = None tmpdir = None try: ... finally: if conn is not None and conn.isconnected(): conn.disconnect() if tmpdir is not None: shutil.rmtree(tmpdir) Good catch, added. It would be nice if this message (both instances of it) included sys.argv[0], so that it is obvious to the user what script is this script: +syslog.syslog(syslog.LOG_ERR, 'Giving up. This script may be safely re-executed.') It is included by syslog: Nov 1 11:13:24 edsel renew_ra_cert: Connection to ldap://localhost:7389 failed, sleeping 30 Yep, but it might be nice to show the full path to the script, since it is not in $PATH. No spaces in kwarg assignment please: +tmpdir = tempfile.mkdtemp(prefix = tmp-) OK. You might want to include sleeping 30 s in this message as well: +except Exception, e: +syslog.syslog(syslog.LOG_ERR, 'Updating renewal certificate failed: %s' % e) +time.sleep(30) Sure, added that. I also added a missing update. I added handling for ldap.SERVER_DOWN as a NetworkError instead of a DatabaseError. rob Honza -- Jan Cholasta ___ Freeipa-devel mailing list Freeipa-devel@redhat.com https://www.redhat.com/mailman/listinfo/freeipa-devel
Re: [Freeipa-devel] [PATCH] 1068 wait for LDAP when renewing the RA
Jan Cholasta wrote:
On 1.11.2012 16:32, Rob Crittenden wrote:
Jan Cholasta wrote:
Hi,
On 24.10.2012 21:24, Rob Crittenden wrote:
All the certs are pretty critical in certificate renewal but the agent
cert has the distinction of having to be updated in multiple places. It
needs to exist in both LDAP servers.
It is possible that one or both of these servers may be down briefly
during renewal so we need to be a bit more robust in our handling. This
will wait up to 5 minutes per server to try to update things, and
syslog
when failures occur.
It is now also safe to re-run this in case something catastrophic
happens. One would just need to manually run this to load the required
data into LDAP.
rob
I believe that there should be a break statement after the updated =
True line:
+updated = True
+except errors.NetworkError:
Sure is, added.
It would be nice if this message said 30 s instead of just 30:
+syslog.syslog(syslog.LOG_ERR, 'Connection to %s failed,
sleeping 30' % dogtag_uri)
Sure.
The continue statement is redundant:
+attempts += 1
+continue
except errors.EmptyModlist:
Yup. I used to have code executed after the try/except/finally. Removed.
The variables you access in both of the finally blocks (conn and tmpdir)
may be unbound. This can be fixed like this:
while attempts 10:
conn = None
tmpdir = None
try:
...
finally:
if conn is not None and conn.isconnected():
conn.disconnect()
if tmpdir is not None:
shutil.rmtree(tmpdir)
Good catch, added.
It would be nice if this message (both instances of it) included
sys.argv[0], so that it is obvious to the user what script is this
script:
+syslog.syslog(syslog.LOG_ERR, 'Giving up. This script may be safely
re-executed.')
It is included by syslog:
Nov 1 11:13:24 edsel renew_ra_cert: Connection to ldap://localhost:7389
failed, sleeping 30
Yep, but it might be nice to show the full path to the script, since it
is not in $PATH.
Ok, added.
rob
From 9071e4b5c6ffb04929c1a68c6cd21359134660ad Mon Sep 17 00:00:00 2001
From: Rob Crittenden rcrit...@redhat.com
Date: Tue, 23 Oct 2012 14:07:13 -0400
Subject: [PATCH] Wait for the directory server to come up when updating the
agent certificate.
It is possible that either or both of the LDAP instances are being restarted
during the renewal process. Make the script retry if this is the case.
It is also safe to re-run this script if it fails. It will take the current
ipaCert certificate and attempt to update the agent information in LDAP.
https://fedorahosted.org/freeipa/ticket/3179
---
install/restart_scripts/renew_ra_cert | 100 +++---
ipaserver/plugins/ldap2.py| 3 +
2 files changed, 72 insertions(+), 31 deletions(-)
diff --git a/install/restart_scripts/renew_ra_cert b/install/restart_scripts/renew_ra_cert
index 14cbc114ca9d38697cdb2a24b41ccf2ac2c66389..048a1ab18825c29d16d62e2a68e4b54dc79f032d 100644
--- a/install/restart_scripts/renew_ra_cert
+++ b/install/restart_scripts/renew_ra_cert
@@ -23,6 +23,7 @@ import sys
import shutil
import tempfile
import syslog
+import time
from ipapython import services as ipaservices
from ipapython.certmonger import get_pin
from ipapython import ipautil
@@ -33,6 +34,7 @@ from ipapython.dn import DN
from ipalib import x509
from ipalib import errors
from ipaserver.plugins.ldap2 import ldap2
+import ldap as _ldap
api.bootstrap(context='restart')
api.finalize()
@@ -53,41 +55,77 @@ except IOError, e:
syslog.syslog(syslog.LOG_ERR, 'Unable to determine PIN for CA instance: %s' % e)
sys.exit(1)
-try:
-conn = ldap2(shared_instance=False, ldap_uri='ldap://localhost:%d' % DEFAULT_DSPORT)
-conn.connect(bind_dn=DN(('cn', 'directory manager')), bind_pw=dm_password)
-(entry_dn, entry_attrs) = conn.get_entry(dn, ['usercertificate'], normalize=False)
-entry_attrs['usercertificate'].append(cert)
-entry_attrs['description'] = '2;%d;%s;%s' % (serial_number, issuer, subject)
-conn.update_entry(dn, entry_attrs, normalize=False)
-conn.disconnect()
-except Exception, e:
-syslog.syslog(syslog.LOG_ERR, 'Updating agent entry failed: %s' % e)
-sys.exit(1)
+attempts = 0
+dogtag_uri='ldap://localhost:%d' % DEFAULT_DSPORT
+updated = False
-# Store it in the IPA LDAP server
-tmpdir = tempfile.mkdtemp(prefix = tmp-)
-try:
-dn = DN(('cn','ipaCert'), ('cn', 'ca_renewal'), ('cn', 'ipa'), ('cn', 'etc'), api.env.basedn)
-principal = str('host/%s@%s' % (api.env.host, api.env.realm))
-ccache = ipautil.kinit_hostprincipal('/etc/krb5.keytab', tmpdir, principal)
-conn = ldap2(shared_instance=False, ldap_uri=api.env.ldap_uri)
-conn.connect(ccache=ccache)
+while attempts 10:
try:
-(entry_dn, entry_attrs) = conn.get_entry(dn, ['usercertificate'])
-entry_attrs['usercertificate'] = cert
+conn = ldap2(shared_instance=False,
Re: [Freeipa-devel] [PATCH] 1068 wait for LDAP when renewing the RA
Jan Cholasta wrote: On 1.11.2012 16:54, Rob Crittenden wrote: Jan Cholasta wrote: On 1.11.2012 16:32, Rob Crittenden wrote: Jan Cholasta wrote: Hi, On 24.10.2012 21:24, Rob Crittenden wrote: All the certs are pretty critical in certificate renewal but the agent cert has the distinction of having to be updated in multiple places. It needs to exist in both LDAP servers. It is possible that one or both of these servers may be down briefly during renewal so we need to be a bit more robust in our handling. This will wait up to 5 minutes per server to try to update things, and syslog when failures occur. It is now also safe to re-run this in case something catastrophic happens. One would just need to manually run this to load the required data into LDAP. rob I believe that there should be a break statement after the updated = True line: +updated = True +except errors.NetworkError: Sure is, added. It would be nice if this message said 30 s instead of just 30: +syslog.syslog(syslog.LOG_ERR, 'Connection to %s failed, sleeping 30' % dogtag_uri) Sure. The continue statement is redundant: +attempts += 1 +continue except errors.EmptyModlist: Yup. I used to have code executed after the try/except/finally. Removed. The variables you access in both of the finally blocks (conn and tmpdir) may be unbound. This can be fixed like this: while attempts 10: conn = None tmpdir = None try: ... finally: if conn is not None and conn.isconnected(): conn.disconnect() if tmpdir is not None: shutil.rmtree(tmpdir) Good catch, added. It would be nice if this message (both instances of it) included sys.argv[0], so that it is obvious to the user what script is this script: +syslog.syslog(syslog.LOG_ERR, 'Giving up. This script may be safely re-executed.') It is included by syslog: Nov 1 11:13:24 edsel renew_ra_cert: Connection to ldap://localhost:7389 failed, sleeping 30 Yep, but it might be nice to show the full path to the script, since it is not in $PATH. Ok, added. rob Thanks. Please also add the conn = None bit to the first loop: while attempts 10: conn = None try: ... finally: if conn is not None and conn.isconnected(): conn.disconnect() and it's ACK. Honza Done, pushed to master and ipa-3-0 rob ___ Freeipa-devel mailing list Freeipa-devel@redhat.com https://www.redhat.com/mailman/listinfo/freeipa-devel
Re: [Freeipa-devel] [PATCH] 1068 wait for LDAP when renewing the RA
On 1.11.2012 16:54, Rob Crittenden wrote: Jan Cholasta wrote: On 1.11.2012 16:32, Rob Crittenden wrote: Jan Cholasta wrote: Hi, On 24.10.2012 21:24, Rob Crittenden wrote: All the certs are pretty critical in certificate renewal but the agent cert has the distinction of having to be updated in multiple places. It needs to exist in both LDAP servers. It is possible that one or both of these servers may be down briefly during renewal so we need to be a bit more robust in our handling. This will wait up to 5 minutes per server to try to update things, and syslog when failures occur. It is now also safe to re-run this in case something catastrophic happens. One would just need to manually run this to load the required data into LDAP. rob I believe that there should be a break statement after the updated = True line: +updated = True +except errors.NetworkError: Sure is, added. It would be nice if this message said 30 s instead of just 30: +syslog.syslog(syslog.LOG_ERR, 'Connection to %s failed, sleeping 30' % dogtag_uri) Sure. The continue statement is redundant: +attempts += 1 +continue except errors.EmptyModlist: Yup. I used to have code executed after the try/except/finally. Removed. The variables you access in both of the finally blocks (conn and tmpdir) may be unbound. This can be fixed like this: while attempts 10: conn = None tmpdir = None try: ... finally: if conn is not None and conn.isconnected(): conn.disconnect() if tmpdir is not None: shutil.rmtree(tmpdir) Good catch, added. It would be nice if this message (both instances of it) included sys.argv[0], so that it is obvious to the user what script is this script: +syslog.syslog(syslog.LOG_ERR, 'Giving up. This script may be safely re-executed.') It is included by syslog: Nov 1 11:13:24 edsel renew_ra_cert: Connection to ldap://localhost:7389 failed, sleeping 30 Yep, but it might be nice to show the full path to the script, since it is not in $PATH. Ok, added. rob Thanks. Please also add the conn = None bit to the first loop: while attempts 10: conn = None try: ... finally: if conn is not None and conn.isconnected(): conn.disconnect() and it's ACK. Honza -- Jan Cholasta ___ Freeipa-devel mailing list Freeipa-devel@redhat.com https://www.redhat.com/mailman/listinfo/freeipa-devel
