7eb0c29 | Daniel Zahn | 13 September 2016, 19:41:58 UTC | monitoring/base: add temperature monitoring via NRPE Adds an Icinga check on all non-virtual hosts, to check the temperature via IPMI and NRPE. The check_ipmi_sensors plugin can check different IPMI sensor types. This limits it to temperature. [-T <sensor type>] limit sensors to query based on IPMI sensor type. Examples for IPMI sensor types are 'Fan', 'Temperature', 'Voltage', ... See the output of the FreeIPMI command 'ipmi-sensors -L' and chapter '42.2 Sensor Type Codes and Data' of the IPMI 2.0 spec for a full list of possible sensor types. The available types depend on your particular server and the available sensors there. Bug: T125205 Change-Id: Iefd4e699302a7adc15537cdbdb71bcaa0dced18c | 31 May 2017, 01:28:32 UTC |
1c6330f | Daniel Zahn | 30 May 2017, 21:47:36 UTC | admin: add herron to ops group Keith Herron is a new Operations Engineer with root access. Bug: T166587 Change-Id: I82d2e65c291081d647108d14fc71f016fa692ef3 | 30 May 2017, 22:08:47 UTC |
13f218c | Daniel Zahn | 30 May 2017, 21:01:28 UTC | admins: create shell account for Keith Herron Creates a new shell account for new ops member Keith Herron. Key as pasted on P5508 / T166587#3302076. UID matching the user in LDAP with @wikimedia.org email. Adding the user to groups will be a follow-up change. Bug: T166587 Change-Id: Ib0b6b75a66a99d41ff2a1116c01e27194fcc33bc | 30 May 2017, 21:28:41 UTC |
71cc1cd | Andrew Otto | 30 May 2017, 20:31:06 UTC | Add ezachte's new ssh key We verified this key over a google video hangout Change-Id: I05668bbe7a0200664b3761f90d51c3a3d8f5aed4 | 30 May 2017, 20:31:06 UTC |
1ceb82e | andrewbogott | 24 May 2017, 17:39:52 UTC | Horizon sudo panel: Add policy checks Change-Id: I8aa4fa042945e2ba5068c6684ea8c493ff4a8b73 | 30 May 2017, 18:34:22 UTC |
f9a7b04 | andrewbogott | 24 May 2017, 15:33:54 UTC | Horizon sudo panel: Better distinguish between 'create' and 'modify' actions - Creation now properly reports if you try to create a duplicate rule - Modify has a read-only 'name' field to avoid accidental renames Bug: T162097 Change-Id: I0fcfd599431c94ee09e7da9389e644e5aba19d0d | 30 May 2017, 18:34:11 UTC |
bd7efd6 | andrewbogott | 10 May 2017, 19:27:54 UTC | Horizon: Add sudo policy panel This works but the error messages are cryptic... we should at least detect naming conflicts and report those failures correctly. Bug: T162097 Change-Id: I383456e8d94bbd9488b09550ef4dc9155e0bc69f | 30 May 2017, 18:33:57 UTC |
858bbf5 | Jaime Crespo | 30 May 2017, 16:10:51 UTC | mariadb: Remove db2048 from the list of full reimages, add db2044 db2048 has just been reinstalled with jessie, db2044 will be next. Change-Id: I42358e1acd478f878d5953b2166fb0b632b46c91 | 30 May 2017, 16:22:53 UTC |
f75600f | Alexandros Kosiaris | 25 May 2017, 09:58:37 UTC | Add jdittrich to analytics-users Bug: T165943 Change-Id: I986f56320c3be0a60826fa5edd172d67db8a6200 | 30 May 2017, 16:13:56 UTC |
9455ff3 | Filippo Giunchedi | 30 May 2017, 16:09:30 UTC | install_server: reimage ms-be2001 with stretch This machine is up for decom and no longer in swift anyway, try stretch reimage. Bug: T162609 Change-Id: Ib68a62816c9888f1ea7511b24e7cf234761e2585 | 30 May 2017, 16:11:10 UTC |
9b07973 | Marko Obrovac | 29 May 2017, 17:37:05 UTC | ChangeProp: Add Redis/Nutcracker connection info Bug: T161710 Change-Id: Iab3c0d43879d79cb39f8e77594f0f110a3d95ebf | 30 May 2017, 14:33:41 UTC |
6b68236 | Emanuele Rocca | 30 May 2017, 11:06:44 UTC | prometheus: add hwmon collector to default set The hwmon collector reports hardware monitoring and sensor data, including thresholds for temperatures to be considered higher than normal. For example: node_hwmon_temp_celsius{chip="platform_coretemp_1",sensor="temp11"} 30 node_hwmon_temp_crit_celsius{chip="platform_coretemp_1",sensor="temp11"} 95 Add it to the default set. Bug: T125205 Change-Id: I8ec5eb8a4f5b99d8f14e0604df136a2773694b1e | 30 May 2017, 14:14:11 UTC |
b853497 | Brandon Black | 25 May 2017, 18:44:46 UTC | interface-rps: optional NUMA awareness Change-Id: I2a65c71644bcaec2ec4d4f025ec45abfa176c69a | 30 May 2017, 14:01:50 UTC |
957297d | Brandon Black | 26 May 2017, 15:14:20 UTC | interface-rps: refactor opts handling a bit Change-Id: I3ddb8adc1a7fe66fcd9663461ae99d7f96853364 | 30 May 2017, 14:00:59 UTC |
b919724 | Brandon Black | 25 May 2017, 18:44:30 UTC | interface-rps: clean up typing/format-string issues Change-Id: I23f76548b5681d60ec692a2ce661d2e6f75f27ca | 30 May 2017, 14:00:39 UTC |
71d5309 | Brandon Black | 25 May 2017, 17:43:02 UTC | interface-rps: update documentation Change-Id: I58a1f17ef602b72c47deed2a45581008f30b7bcf | 30 May 2017, 14:00:07 UTC |
826c4d6 | Alexandros Kosiaris | 30 May 2017, 13:13:28 UTC | calico: Organize the required per DC hieradata Split the DC specific configuration in the respective locations. Also set the codfw configuration correctly Change-Id: I7ab562c9d8c0f176a93a437bc5f6b491e53236b0 | 30 May 2017, 13:13:28 UTC |
99a1a13 | Moritz Muehlenhoff | 18 May 2017, 08:54:22 UTC | Also strip rpcbind/nfs-common deps on jessie installs On a freshly installed jessie system we have two packages which got pulled in during base install by nfs-common/rpcbind, so strip these as well: root@kubetcd2001:~# apt-get autoremove Reading package lists... Done Building dependency tree Reading state information... Done The following packages will be REMOVED: libnfsidmap2 libtirpc1 Bug: T106477 Change-Id: I871a4530fa5164daba5a131425c619b8a2415104 | 30 May 2017, 12:53:20 UTC |
227c20c | Filippo Giunchedi | 30 May 2017, 10:04:37 UTC | hieradata: set max_body_size for swift::proxy Bug: T166482 Change-Id: Ie4df1817f5bd1bc031582c12af1de44d32590fec | 30 May 2017, 10:46:38 UTC |
5607a3d | Filippo Giunchedi | 30 May 2017, 09:53:31 UTC | tlsproxy: add support to change max_body_size mediawiki to/from swift communications now happen on https and thus through nginx, we'll need to bump the max body size nginx will let through. Bug: T166482 Change-Id: Ied6a8ef5600925e7621380db97ddf8a551411f09 | 30 May 2017, 10:43:44 UTC |
4b37c7d | Alexandros Kosiaris | 30 May 2017, 10:13:58 UTC | calico-node: Enable IPv6 support For calico to enable IPv6 support the IP6 variable needs to be defined to a valid IPv6 address and in fact the one present on the host. Use the fact ipaddress6 for that Change-Id: I6c2b230cee52ddfb2b276648a331e52b3673c7f7 | 30 May 2017, 10:26:28 UTC |
1adfab1 | Faidon Liambotis | 29 May 2017, 21:50:54 UTC | Blacklist the parallel port kernel modules stretch's kernel seems to be emitting a warning in QEMU virtual machines (Debian bug #857624) and we don't use nor need parallel ports anyway :) Change-Id: I916da9d05ef5db7c9ab46749ad930c1ff759734b | 30 May 2017, 10:10:20 UTC |
c1abc0e | Guillaume Lederrey | 29 May 2017, 14:33:04 UTC | elasticsearch - correct naming of curator config files Bug: T166154 Change-Id: Ib2099494ed2cc0203e6d5a67cb0f3e8c0d35eb5a | 30 May 2017, 09:29:44 UTC |
ef76ec6 | Guillaume Lederrey | 29 May 2017, 16:26:59 UTC | elasticsearch - configure logging for elasticsearch-curator curator logs are sent to /var/log/elasticsearch/curator.log. All logs in /var/log/elasticsearch are already managed by logrotate, no need for additional configuration. Bug: T166154 Change-Id: Ie4d6b02267fd616b84dee7b1341e32fe92251e7c | 30 May 2017, 09:27:06 UTC |
ce19f12 | Filippo Giunchedi | 29 May 2017, 13:23:55 UTC | Don't symlink systemd service instances In stretch systemctl will explicitly refuse to operate on symlink'ed units. For templated services the symlink isn't necessary because systemctl already DTRTs Bug: T166389 Change-Id: I034eb945189f3aec693a15994e9b29e1b9dcb2a6 | 30 May 2017, 09:05:31 UTC |
0fb1c3f | Guillaume Lederrey | 29 May 2017, 19:19:29 UTC | elasticsearch - ignore some warnings related to 5.3.2 upgrade This actually fixes a configuration error. Bug: T163708 Change-Id: I33a30ef46b88e76f4b3039c549097aa117f366c8 | 29 May 2017, 19:20:02 UTC |
c5b262a | Riccardo Coccioli | 29 May 2017, 17:12:56 UTC | raid/hpssacli: allow NRPE to execute all commands Bug: T163998 Change-Id: Icb11ea112db468986c4aac48defc1dc7efbd21f5 | 29 May 2017, 17:12:56 UTC |
368f49a | Guillaume Lederrey | 10 May 2017, 16:44:56 UTC | elasticsearch - silence some loggers for elastic 5.3 upgrade The ParseField deprecation will be fixed by https://gerrit.wikimedia.org/r/#/c/353100, the RestController deprecation needs an Elastica upgrade. Pending those fix, we need to hide those deprecations. Change-Id: I37b4368d15c6f1e931d24dae00531f9105066c0d | 29 May 2017, 17:05:12 UTC |
09bb665 | Faidon Liambotis | 17 May 2017, 00:55:12 UTC | raid/hpssacli: check for cable errors/no batteries T163777 had a case where the following "show status" resulted in an OK: Controller Status: OK Cache Status: Permanently Disabled We already fixed the check to emit a WARNING when Cache Status is not "OK" or "Not Configured" in a previous commit, but it seems there's another thing we could check: "controller slot=N show detail". This had a few more values we could check, and specifically: Cache Status Details: Cable Error Battery/Capacitor Count: 0 Emit CRITICAL for the former, and CRITICAL for the latter if the count is 0 and the argument --no-battery hasn't been passed to us. This is untested on systems that have no battery by design -- hopefully, Cable Error won't be reported on these. Bug: T163998 Change-Id: Iaa099ec825a86445f6e79cfad895e9aec757725c | 29 May 2017, 16:37:33 UTC |
71d8691 | Faidon Liambotis | 17 May 2017, 00:26:26 UTC | raid/hpssacli: WARN on permanently disabled cache Commit 94f371f1f400a636694eead5a228285314a48db3 skipped the Cache Status line of "controller slot=N show status", in an effort to make it not warn on HP SSD Smart Path configurations where Cache Status was set to "Not Configured". We were checking instead the LD Acceleration Status for each logical drive, which was indicative in the past. We have now seen a case out in the wild where Cache Status is set to "Permanently Disabled", indicative of an issue with the hardware's battery, but LDs all report "LD Acceleration Status: Controller Cache" and "Caching: Enabled". Change the Cache Status check to explicitly check for "OK" or "Not Configured" and otherwise emit a WARNING. Bug: T163998 Change-Id: Ib186044a8d349ca82f112c4f90da0ca20ccad96f | 29 May 2017, 16:37:33 UTC |
7bdea41 | Faidon Liambotis | 29 May 2017, 13:30:56 UTC | pmacct: update config, push output to Kafka Update our pmacct (nfacctd) config, to be able to push to a Kafka topic with the name of "netflow". This is in turn used by the Druid pipeline (via Tranquillity) and eventually shown into Pivot dashboards. While at it, remove the GeoIP stuff that we weren't using and set bgp_peer_as_skip_subas to true. Also set bgp_aspath_radius to 2, which is lossy (trims the AS path to 2, losing the rest) but which OTOH gives us a better visibility of the amount of traffic exchanged with the hop right after our transits. Change-Id: I3892851c723eb6c0c29fc6df8feb606e0ca88c43 | 29 May 2017, 16:32:37 UTC |
bb43a2a | Faidon Liambotis | 29 May 2017, 12:57:00 UTC | pmacct: fix firewall rules We want routers to be be able to communicate to pmacct (netflows and BGP), but router loopbacks aren't in $PRODUCTION_NETWORKS which is what the firewall rules currently allow. Add a list of loopbacks per site, hardcoded within the pmacct manifest. Adding it to the network data Hiera was considered but rejected as an option, since it would fit poorly into that structure and (depending on where they would be added) open up connections to random services from the networking equipment. Hardcoding it into pmacct should be OK for now, and we can think about factoring it out when we'll need to reuse this information elsewhere. Change-Id: I898b7df9d502b842e00676e6efd5a0b2a0467669 | 29 May 2017, 16:30:07 UTC |
34130f5 | Faidon Liambotis | 29 May 2017, 16:18:48 UTC | aptrepo: fix elastic.co's update filter The current grep-dctrl hook emits this error: grep-dctrl: inconsistent modifiers of simple filters. Fix this by moving -e in the sub-filter and only in the first one where it's actually needed. Change-Id: I6e8c9642f62f11042d21b442ba2e0ecf4cb1038a | 29 May 2017, 16:23:58 UTC |
124142a | Giuseppe Lavagetto | 29 May 2017, 15:59:14 UTC | profile::nutcracker: correctly handle monitoring, refactoring * Add a monitoring_port parameter to set which port to monitor, if any * Add base settings for both redis and memcached, allowing only to override those, allowing a more DRY approach. * Add a parser function to merge the hashes, this could be avoided with either the future parser or a better base class. Change-Id: I35ff9fecc6b342de640e6e37e5b7764797080603 | 29 May 2017, 16:20:37 UTC |
161cb67 | Moritz Muehlenhoff | 29 May 2017, 11:39:02 UTC | Remove obsolete mediawiki::migrate class terbium reimage is complete Change-Id: I27e37985c4d6255840ab812d7c0ff0f1675f52c9 | 29 May 2017, 15:53:12 UTC |
55d7e5d | Giuseppe Lavagetto | 29 May 2017, 14:58:01 UTC | role::scb: fix redis host list for changeprop Change-Id: Ied6c0c8e76ceb1418f63565fd6ca6bf0d2ff0716 | 29 May 2017, 14:58:01 UTC |
7988dea | Giuseppe Lavagetto | 29 May 2017, 14:40:14 UTC | role::scb: add nutcracker for changeprop Also add a more generic profile::nutcracker to use in the general case. Change-Id: I910c095c46ccacaf9f9e52699d9c7373a5a3fb07 | 29 May 2017, 14:50:45 UTC |
7395217 | Alexandros Kosiaris | 29 May 2017, 14:21:49 UTC | Specify the correct service IPs for kubernetes clusters eqiad and codfw were mistakenly swapped Bug: T165732 Change-Id: I491b278a0643d2b3ccbdc0594568eb50867d9e08 | 29 May 2017, 14:23:28 UTC |
f383987 | Alexandros Kosiaris | 29 May 2017, 13:25:36 UTC | Utilize the allocated service ips in kubernetes Since we now have allocated service IP ranges in T165732, utilize them. While these IPs are expected to never be found on the wire due to all the NATs done by kubernetes, it is still considered good to allocate some IPs we know and allocate instead of having 192.168/16 IPs configured there for uniformity purposes Bug: T165732 Change-Id: I68982912dce8ddc659d76029bd9078312f7e068c | 29 May 2017, 13:51:54 UTC |
09265a7 | Guillaume Lederrey | 29 May 2017, 12:28:10 UTC | elasticsearch - introduce curator configs to enable / disable replication Bug: T166154 Change-Id: I9a074b31fba7764838a6a7114c47ebc1110e59d3 | 29 May 2017, 12:52:52 UTC |
07e4a05 | Alexandros Kosiaris | 29 May 2017, 12:17:37 UTC | Kubernetes: Add IPv6 mapped addresses While these boxes are not required to have IPv6, it is prudent to start usage having it configured nicely in order to test and find limitations if any Change-Id: I718748cf52c11b70ae7eb3a111676749545bba5b | 29 May 2017, 12:17:37 UTC |
8d9224f | Emanuele Rocca | 29 May 2017, 10:05:54 UTC | debug_proxy: increase client_max_body_size The default maximum client request body size is 1M, bump it to 100M. Bug: T165324 Change-Id: Ia4c372a3ccaa66a37bf5d6adc6014166567d1ae4 | 29 May 2017, 10:05:56 UTC |
c9bdc24 | Filippo Giunchedi | 16 May 2017, 15:08:43 UTC | prometheus: report puppet agent stats Introduce prometheus::node_puppet_agent to parse puppet state directory and report status (disabled, failed, etc) as metrics. Change-Id: I0eeb80cf68addacdd651b8eb80c159c711498a1f | 29 May 2017, 09:31:45 UTC |
f1a6942 | Moritz Muehlenhoff | 29 May 2017, 09:15:11 UTC | Record extended account expiry date for piccari Change-Id: I9df9963a04e86b5c820339aead861be9e14819a9 | 29 May 2017, 09:15:11 UTC |
578ba51 | Antoine Musso | 25 May 2017, 20:54:16 UTC | Fix spec for various modules base::service seems to now invoke logstash::conf, add 'logstash' to the list of fixtures. nrpe specs were failling with 'unknown init system'. Set the fact to 'systemd'. Change-Id: I0a4e9c41329bf71c9a5dc11ef3c033fb8ce0c74f | 29 May 2017, 09:08:31 UTC |
fe9f9f1 | Jaime Crespo | 29 May 2017, 08:35:01 UTC | mariadb: Increase the binlog_cache_size to 10M There seems to be very large transactions, that go over the 1M limit and cause slowdown due to the binlog having to be written into disk: https://grafana.wikimedia.org/dashboard/db/mysql?orgId=1&var-dc=codfw%20prometheus%2Fops&var-server=db2049&panelId=19&fullscreen&from=1495793965043&to=1495974377121 Increase the size 10x (from 1M to 10M to accomodate for those large transactions). The transaction size, however, should go down. Change-Id: I9c834a64cfe9b35ab40f58c419fd4fdcd15f8084 | 29 May 2017, 08:46:55 UTC |
ab66323 | Jaime Crespo | 26 May 2017, 14:16:24 UTC | mysql-client: Install colordiff (neodymium & sarin) The colordiff utility will make easier to detect differences on sql output, one of the things done on the client servers. diff --color was only added on diffutils 3.4 (debian >= stretch). Change-Id: I600b582ce5089c23630cd93ad0ee82cc03dfb989 | 29 May 2017, 08:45:01 UTC |
bc525a9 | Filippo Giunchedi | 29 May 2017, 08:30:09 UTC | decom ms-be2001 - ms-be2012 Leave systems as role::spare::system to get puppet updates until physical decom happens. Bug: T162785 Change-Id: I5546db2058c15299d4c87b0d5e322744417d1d20 | 29 May 2017, 08:37:33 UTC |
62169d8 | Manuel Arostegui | 29 May 2017, 07:32:36 UTC | mariadb: Decommission db1023 Bug: T166486 Change-Id: I2effec0d29d6b8c79e70bb49e65825eee52b21cc | 29 May 2017, 08:33:57 UTC |
61e66ef | Riccardo Coccioli | 27 May 2017, 22:10:14 UTC | Monitoring: remove spaces from list of interfaces - When generating the list of interfaces in check_eth, there are several whitespaces left for all the interfaces filtered. This clean the list without leaving whitespaces. Bug: T166372 Change-Id: Ie642af5146b0273a5fa92d74218029ebcce1d915 | 29 May 2017, 07:52:59 UTC |
5629a48 | Giuseppe Lavagetto | 24 May 2017, 08:59:29 UTC | role::kubernetes::worker: upgrade calico everywhere Bug: T165024 Change-Id: Iaeea2f641a162908bd637aac126814277fb4da77 | 28 May 2017, 15:25:05 UTC |
08e09a2 | Daniel Zahn | 27 May 2017, 17:31:09 UTC | admins: revoke ladsgroups key temporarily Change-Id: I7fe7544fefbd2cf24be827e965480332ca4ce96a | 27 May 2017, 17:33:27 UTC |
8affb59 | Daniel Zahn | 26 May 2017, 23:12:00 UTC | phabricator: reactivate exim ganglia stats I had this commented as a stop-gap for the spam-from-labs issue we had recently. Since the mail alias on the affected labs instance has been fixed and we don't have a replacement for these graphs yet, re-activate it as it was before. Change-Id: I6aa730330cbb3f6846b65723db1a8f7a566bb37f | 26 May 2017, 23:14:52 UTC |
13b9791 | Daniel Zahn | 26 May 2017, 04:40:54 UTC | wikistats: install php7.0-xml if on stretch Bug: T165879 Change-Id: I03f04886dbdcfd616cd199d5fd2b5131f76fdd93 | 26 May 2017, 19:07:05 UTC |
535989e | Daniel Zahn | 26 May 2017, 17:22:35 UTC | planet: add Wikimedia Cloud Services blog feed Change-Id: I781259e180976709f4226b52fac17fdf37dd291a | 26 May 2017, 17:52:48 UTC |
3778bdf | Giuseppe Lavagetto | 26 May 2017, 11:48:46 UTC | jobrunner: bump up the number of htmlCacheUpdate jobs temporarily We have a long backlog right now, and room for growth in terms of used resources. Change-Id: I9968c2c8d5401e0ca08285d91bd86abe3d431c0e | 26 May 2017, 11:48:46 UTC |
becb413 | Giuseppe Lavagetto | 26 May 2017, 10:22:10 UTC | profile::redis::multidc_instace: fix template Change-Id: Iebe4d73723a5d6b1ec0af2081ccd285c043d4cce | 26 May 2017, 10:22:10 UTC |
bddae83 | Giuseppe Lavagetto | 26 May 2017, 08:11:05 UTC | role::jobqueue_redis: add redis instances for changeprop Since we don't need replication given CP is active-active, also allow to define replica: false in redis::shards in order to control such a case. For now this instances will be used for RB automatic blacklisting Bug: T161710 Change-Id: I4ce64011e546cb94ad104e7abeee39ffc4c0f301 | 26 May 2017, 10:10:17 UTC |
61b1893 | Jcrespo | 19 May 2017, 10:41:33 UTC | Revert "Revert "mariadb: allow reimage of db2048 for upgrade to jessie"" This reverts commit 8e26212a25f7233306eb4136edf4629a3065806a. Change-Id: I4e1fadc7c4352cecf4ec677bd0793d622a66c4aa | 26 May 2017, 09:44:06 UTC |
32903b5 | Manuel Arostegui | 26 May 2017, 07:48:40 UTC | check_private_data_report: Add Jaime's email Change-Id: Ic2e5fe913d1bbf592909c8c7350d6a4528d8aff4 | 26 May 2017, 08:57:50 UTC |
1a7233b | Faidon Liambotis | 26 May 2017, 08:24:31 UTC | check_cpufreq: various cleanups (typos, shell etc.) Change-Id: Iac6570e9ee2cc4c5b8720352e8d8b165625c0661 | 26 May 2017, 08:31:35 UTC |
4a079a1 | Jaime Crespo | 26 May 2017, 08:18:46 UTC | install_server: Reimage db2049 as jessie Bug: T165739 Change-Id: Id952289d96750b6c0250c567eed2685786539ca6 | 26 May 2017, 08:18:46 UTC |
cad024c | Jaime Crespo | 25 May 2017, 08:53:29 UTC | jynus-vimrc: Disable mouse input & enable syntax highlighting Say no to stretch defaults! Change-Id: Ib28cc8f9f778dd967abd1476b74762f848902d42 | 26 May 2017, 08:06:51 UTC |
e755fca | Daniel Zahn | 25 May 2017, 20:52:00 UTC | wikistats: ensure Apache PHP module is installed before site With the current setup puppet runs and libapache2-mod-php-* gets installed but the PHP site won't work until a manual 'dpkg-reconfigure libapache2-mod-php7.0'. This is because it gets installed before the rest of the Apache packages get pulled in. Add missing Require to fix that, also move things around a bit and make more obvious how the PHP version influences package names. Change-Id: Idb4205bee82051a36c3b22588a2fce98c975e89f | 25 May 2017, 21:31:51 UTC |
88b1e1d | Andrew Otto | 17 May 2017, 09:32:27 UTC | Add kafka_version parameter, s/java_packaage/java_home/ in confluent::kafka::client This allows for configuration of which Java is used, and which Kafka version will be installed. Change-Id: I577ed3f0fd9c95ba7ba5df1b49ce1e1680754dfb | 25 May 2017, 19:39:12 UTC |
9c86a4c | Andrew Otto | 23 May 2017, 18:15:08 UTC | Update kafka.sh wrapper script for Kafka 0.10+ This is safe to be merged now, but it will break some uses of the kafka wrapper script in older Kafka versions. Specifically, kafka console-consumer will not work, as it requires --zookeeper in the older Kafka version, but the newer one expects --bootstrap-server. If you need to use the console-consumer, you can bypass this wrapper script to use /usr/bin/kafka-console-consumer and manually pass --zookeeper $KAFKA_ZOOKEEPER_URL directly. Bug: T166164 Change-Id: Ibd24d7f7dd1d897493ed6a61db4704b72954aa25 | 25 May 2017, 18:51:18 UTC |
42d7e37 | Daniel Zahn | 24 May 2017, 17:26:49 UTC | phabricator: set alt_host in redirector to "phab" instead of "fab" fab.wmfusercontent.org does not exist in DNS. phab.wmfusercontent.org does. Code in the redirector has: if ($_SERVER['HTTP_HOST'] !== '<%= @phab_host %>' && $_SERVER['HTTP_HOST'] !== '<%= @alt_host %>') { .. redirect .. So this means redirects also happen if HTTP_HOST is actually phab.wmfusercontent.org, but that is not intended. Change-Id: Ic36ea7a9664a9e1f003cefdfe6170cb8210a4f9d | 25 May 2017, 16:53:51 UTC |
e9958ac | Brandon Black | 25 May 2017, 14:34:12 UTC | r::c::perf - +rxring and +txqueuelen These values have proven useful on 10G bnx2x cards in LVSes, copying over to the caches as well. txqueuelen bump - can only be a good thing (or at worst neutral) under BBR/FQ control. rxring bump - All current text/upload caches show at least some count of receive overruns (the only zeros are on the lower-traffic clusters). At present, all live cache boxes as well as the known upcoming new installs in ulsfo+asia have bnx2x as primary in eth0, so the config basically assumes this for now, until some point in the future when that changes. Change-Id: Id1514abef02ad08a4a982d5c79883907434ae47f | 25 May 2017, 15:50:19 UTC |
918cb40 | andrewbogott | 25 May 2017, 02:12:45 UTC | Horizon: Add some local_settings for ldap access Change-Id: I15307dbf4a9f489ad95734ca3779bc0d79834288 | 25 May 2017, 02:14:08 UTC |
6da3083 | Giuseppe Lavagetto | 17 May 2017, 11:14:18 UTC | role::aqs: use profile::cassandra Also add a switch to profile::cassandra to allow opening connections to the analytics network in case of need, and fix handling of the case we have no TLS encryption. Change-Id: I727779606ff6c83f95e2dd308df3fa34f14e3f77 | 25 May 2017, 12:00:05 UTC |
39b64f2 | Giuseppe Lavagetto | 24 May 2017, 08:56:28 UTC | role::kubernetes::worker: upgrade calico on one host Let's first test the upgrade on one host only. Bug: T165024 Change-Id: I9866a3148a298f5d0e6501b1781fc5aa0b8816df | 25 May 2017, 10:26:25 UTC |
6d160c6 | Daniel Zahn | 24 May 2017, 23:30:41 UTC | admins: add kaldari to analytics-privatedata-users Access to the Hadoop Cluster ("Data Lake") via host stat1002, as requested on ticket. Bug: T166165 Change-Id: I373087f36e00128306af6821cddd94ffe790262c | 25 May 2017, 09:54:58 UTC |
6a5a14e | Giuseppe Lavagetto | 24 May 2017, 08:54:46 UTC | calico: add new version 2.2.0 Bug: T165024 Change-Id: I667d376f151c33e86dfab3de546a9ee502778e66 | 25 May 2017, 07:07:43 UTC |
24b5e2d | Daniel Zahn | 25 May 2017, 00:56:56 UTC | phab: comment out include of exim4::ganglia This caused LOTS of cron spam since today. The chain of events was like: < paladox> !log phabricator upgrading mariadb to 10.2 -> /usr/lib/x86_64-linux-gnu/libmysqlclient.so.18: symbolic link to libmariadb.so.3 -> /usr/sbin/exim: /usr/lib/x86_64-linux-gnu/libmysqlclient.so.18: no version information available (required by /usr/sbin/exim) -> /usr/local/bin/exim-to-gmetric" fails But this cron is running every minute and then tries to send mail to root. /etc/aliases says that root should be root@wikimedia.org and then when it _tries_ to mail that sender verification fails: 550-Verification failed for <root@phabricator.phabricator.eqiad.wmflabs>" which then causes "mail delivery failed" messages which get delivered to all of ops. Since Ganglia is deprecated and i wanted to find a quick stop gap before talking about this some more, i am just commenting that class right now. Change-Id: Iff1ce9cdd059a4239309ab93d367e1bfa790b91d | 25 May 2017, 01:06:20 UTC |
2760d4c | Daniel Zahn | 10 May 2017, 17:16:17 UTC | authdns::server: move 'include standard' to role Change-Id: Ic63b1f5fdffd81a1df0153d9c6054f0f7dd6a591 | 25 May 2017, 00:13:44 UTC |
a6a2ca2 | Daniel Zahn | 10 May 2017, 17:08:19 UTC | dnsrecursor: move 'include standard' to role Change-Id: I6fa652760aeafc846e6ab8541449f41f42b621a4 | 25 May 2017, 00:02:07 UTC |
82af211 | Daniel Zahn | 08 May 2017, 19:50:08 UTC | confd: use logrotate::conf for logrotate Change-Id: I772ffaa22542ca42b4bcd763895e562de7954e30 | 24 May 2017, 23:44:44 UTC |
a79acf1 | Paladox | 08 May 2017, 21:34:09 UTC | Gerrit: Remove "" around T\\d+ in gerrit.config When i did a gerrit upgrade and ran this java -jar gerrit.war init -d review_site It removed the quotes around match = "T\\d+" But in the puppet repo we have quotes around it so puppet re added them back - match = T\\d+ + match = "T\\d+" Thats a puppet run so it added "" Change-Id: If2c073e5f6bad030672f5eba88156844cdee9b6a | 24 May 2017, 21:00:08 UTC |
89b277c | pt1979 | 24 May 2017, 20:38:20 UTC | DHCP/partman: Add DHCP and partman entries for ores200[1-9] Bug:T165170 Change-Id: I2bca295a6baf0fbbe588be6e8fc00d5ea36ebbf1 | 24 May 2017, 20:42:08 UTC |
837db31 | Andrew Otto | 24 May 2017, 19:13:37 UTC | Use is_not_bot filter function for eventlogging mysql consumer Bug: T67508 Change-Id: Ib5156dcf582c9829e336e4c58fe7184cc4e590d2 | 24 May 2017, 19:13:37 UTC |
269f8dd | Ottomata | 24 May 2017, 17:41:47 UTC | Revert "Changes needed for upgrading to Druid 0.10" We have to roll back to 0.9. Hadoop indexing jobs don't work due to java 8 vs java 7 issues. This reverts commit 6fbadbad2658c964e46203dd447707676016f83e. Change-Id: Ide4e009ca1b4534649462f44aa823e356bf89a6a | 24 May 2017, 17:42:31 UTC |
1fefeeb | Brandon Black | 24 May 2017, 17:03:17 UTC | r::c::perf - FQ outbound flow rate cap @ 1Gbps This should help clamp down brief rate spikes in local intercache and cache:app flows (aka TCP conns) to no more than 10% of interface bandwidth, and throttle some of the higher-traffic remote proxies (e.g. opera, zscaler, google, etc) to more-reasonable fractions of our available outbound host and transit bandwidth as well. Bug: T147569 Change-Id: If26405f59d6646842e7d6d459eb1a12a046c179b | 24 May 2017, 17:07:35 UTC |
9403053 | andrewbogott | 12 May 2017, 22:45:51 UTC | novastats: Update some reports to use more up-to-date code. The old 'novastats.py' library is no longer needed since mwopenstackclients does most of the querying we want. This patch replaces some old scripts with a few newer example scripts that use mwopenstackclients. Change-Id: I8e054c973dcdaa731688788b5087043308779f96 | 24 May 2017, 16:12:23 UTC |
fef24b7 | andrewbogott | 12 May 2017, 21:44:19 UTC | openstackclients: add an optional project arg to allinstances() Change-Id: I5b18939213dc9a1d4879c1309ff14664e22755a5 | 24 May 2017, 16:07:39 UTC |
fd9a126 | Moritz Mühlenhoff | 24 May 2017, 14:41:50 UTC | Extend expiry dates for two accounts Change-Id: I66a8ffd574fa5cfdf3482438f9cb1b52b6920d7a | 24 May 2017, 14:41:50 UTC |
aedd882 | Bryan Davis | 20 May 2017, 20:46:53 UTC | tools: have maintain-kubeusers chown $HOME/.kube Chown the $HOME/.kube directory to the tool account. Also protect the generate credentials with chattr. Bug: T165875 Change-Id: I37bc5517ab4bd8b646e0c63d9140c1cc2633e9ea | 24 May 2017, 14:20:38 UTC |
7b4e9f8 | Riccardo Coccioli | 24 May 2017, 10:48:19 UTC | Puppet: more reliable run-puppet-agent - when using the --failed-only option, consider last run failed if the summary yaml file is not valid. Change-Id: I886646b4a250b5e7ca30804837d4d73d6fa7dda2 | 24 May 2017, 14:02:30 UTC |
ed6a7e5 | andrewbogott | 19 May 2017, 22:30:14 UTC | Tidy up tools node motd Change-Id: I795ceb01adf3d0e6b0d839b88530ca7b75f840b2 | 24 May 2017, 13:58:32 UTC |
6fbadba | Andrew Otto | 03 May 2017, 18:55:19 UTC | Changes needed for upgrading to Druid 0.10 Druid 0.10 depends on Java 8. The package should install it as a dependency when installing, but it will not be the default Java. Setting JAVA_HOME explicitly in daemon env.sh files. Bug: T164008 Change-Id: I83a36bdfbcbe13f1c06471446a9bc5a0aa3e0941 | 24 May 2017, 13:43:37 UTC |
9d0bfe7 | Alexandros Kosiaris | 24 May 2017, 11:36:14 UTC | logrotate: Fix uwsgi postrotate script Using invoke-rc.d on jessie systems is plain wrong. For now using the service command wrapper and reevaluate in the future if we should switch to systemctl Change-Id: I584532cf9f8c25d858b3035b78de2a3163fdc25e | 24 May 2017, 11:39:14 UTC |
aa116a7 | Alexandros Kosiaris | 24 May 2017, 11:22:47 UTC | Specify the correct host for wikidata icinga config Since we changed it to www.wikidata.org, we also need to change the services referencing it Change-Id: I5633a0010ffef232d991de52126c1f3486767d56 | 24 May 2017, 11:24:12 UTC |
2c7089d | Alexandros Kosiaris | 24 May 2017, 10:31:58 UTC | Decouple wikidata monitoring from the IP address Using the IP address of text-lb.esams was plain wrong as it would alert even when esams was depooled. Change the definition to use the FQDN, quality the definition correctly and stop using the -I on the command but rather rely on the correct DNS resolution done by -H of check_http Change-Id: I2efa53ea6dc8677cdc3b4769dfdb26a81b09cd33 | 24 May 2017, 10:53:19 UTC |
01bff75 | Moritz Mühlenhoff | 24 May 2017, 10:44:24 UTC | Revert "Use gdb from jessie-backports on jessie" This reverts commit 026c81a83071fca6a16939e3a405d0487e00b8c1. Change-Id: Ia159081fed2e7e6e3e438a85c930d8af386ae710 | 24 May 2017, 10:44:24 UTC |
026c81a | Moritz Muehlenhoff | 22 May 2017, 13:36:50 UTC | Use gdb from jessie-backports on jessie Fixes access to TLS variables in threaded programs (in particular HHVM) (among other changes in four major releases). It's a low level debugging tool, so shouldn't have any impact on existing setups. Change-Id: I7c7540660368846bd9486c83208f0032e4974b49 | 24 May 2017, 10:40:30 UTC |
ab02b19 | Emanuele Rocca | 24 May 2017, 08:47:24 UTC | prometheus: enable qdisc collector on cache hosts Enable queueing discipline (qdisc) metric collection on cache hosts. Traffic control statistics are useful, among other things, to monitor BBR's behavior. Bug: T147569 Change-Id: I490f72235fd873903fdf1baf322cda602d21ae0a | 24 May 2017, 10:25:53 UTC |
7a3cb90 | Antoine Musso | 24 May 2017, 10:04:31 UTC | beta: set profile::etcd::tlsproxy::read_only=false 79cfdefd50 added a new setting so we can switch etcd readonly whenever doing a switchover (T159687). In production the setting is applied via the hieradata role hierarchy: hieradata/role/common/configcluster.yaml hieradata/role/eqiad/configcluster.yaml However on deployment-prep the role hierarchy is not looked up causing on deployment-etcd-01: Could not find data item profile::etcd::tlsproxy::read_only in any Hiera data file... Set the value to false, assuming on beta we want etcd to be writable. Change-Id: I99f0bf11112de2a81bbbf131ece01eaea7871227 | 24 May 2017, 10:04:31 UTC |
776b1d5 | Jaime Crespo | 23 May 2017, 17:03:09 UTC | raid: Implement the option to check write cache policies Enable all databases to enforce WriteBack policy, though an error (not a warning) if they go down to WriteThrough- they probably will cause performance problems and have something anomalous like a BBU problem or something else. Ignore those alerts on the handle-raid scripts to avoid creating failed disk tasks. This currently only works on megacli systems, although the plan is to deploy it to all systems where the feature makse sense (megacli are the ones that are older and cause daily issues). Bug: T166108 Change-Id: I36fc6fb115c2d9e9d88391ea5a9230d6389f781c | 24 May 2017, 09:17:05 UTC |
b182709 | Gilles Dubuc | 15 March 2017, 10:23:37 UTC | Enable memcache-based Thumbor broken thumbnail throttling Bug: T151065 Makes memcache run on the thumbor machines, as well as nutcracker to pool connections going to both. This colocation makes sense because the memcache use is minimal and thumbor's throttling feature is tolerant of memcache being down. Change-Id: Idc5f10324e1c0877901393949acccc618b340dde | 24 May 2017, 08:34:16 UTC |
ef3e76b | Riccardo Coccioli | 23 May 2017, 22:05:08 UTC | Puppet: run-puppet-agent improvements - fix bug when --failed-only is set - do not imply --quiet with --failed-only Change-Id: Iaf441764d3893f560cbcc330a182d2eb2ad264ce | 23 May 2017, 22:11:30 UTC |
92b5ba2 | Brandon Black | 23 May 2017, 21:25:21 UTC | r::c::perf - raise fq flow_limit to 300 Still seem some rare flow_plimit drops, generally on a single queue and only affecting ~20% of cache boxes. Hopefully this will make them truly-rare (there will probably always be bad edge cases where these drops are the most appropriate course of action). Bug: T147569 Change-Id: Ib0b19da2cdd53f8b8928165d770c49f16a6126bc | 23 May 2017, 21:26:41 UTC |