Author name: scott

On Tuesday, March 5, 2013, we will have a scheduled downtime from 6:00am to 7:00am EST.

UPDATE: This downtime did not happen and the work has been postponed to March 19, 2013.

Who is affected:

• Users of the public login machines (soak, wash, rinse, spin, tux, opus), certain CS websites (jobs, kiosk, msdnaa, search, wiki), database servers, runscript, CAS, lpdrelay (printing), labpc machines (Friend Center fishbowl lab) and the ionic cluster.

What is happening:

• These hosts will receive a critical security update and be rebooted.

Why is it happening:

• There is a critical kernel security patch available for our Springdale hosts. It addresses a specific security vulnerability. As this is a kernel update, all machines must be rebooted. Actual expected downtime should only be a minute or two for each host to reboot.

The temperature in the central computing facility has risen very quickly to critical levels this morning. OIT and facilities are aware of the problem. We have already started to see equipment failures. To protect our infrastructure, we are very likely to begin shutting down equipment.

UPDATE: At 10:43am, OIT reports, \”All air-handlers at the HPCRC are currently stopped.\”

UPDATE 10:55am: The penguin machines (tux, opus) and cycles (wash, rinse, spin, soak) are being shutdown. Our primary DNS server automatically shutdown due to the high temperature.

UPDATE: At 10:56am, OIT reports, \”The air handlers are running again and temperatures are returning to normal.\” CS Staff is monitoring temperature sensors in our area.

UPDATE 11:13am: Primary DNS server has been restarted. We expect to restart penguins and cycles within 15 minutes. The switch for the ionic cluster failed. We have a spare in the CS building. It will be a while (no ETA yet; could be a full day) before it is configured, installed, and operational.

UPDATE 2:00pm: We may not be out of the woods yet. We are seeing elevated temperatures again and we are keeping a close eye on our systems. From OIT, \”HPCRC control system is having problems again. Facilities staff are still at HPCRC and are responding.\”

UPDATE 2:55pm: We have seen temperatures peak and then decline again. OIT now reports, \”HPCRC cooling is now operating normally.\”

UPDATE 4:20pm: Facilities believes that they have identified and corrected the root cause. As a precaution, facilities staff will stay at the HPCRC all night to be able to respond quickly in the event of another problem.

On Saturday, February 2, 2013, from 5:00am to 9:00am, the campus network will be offline for scheduled work.

See: http://helpdesk.princeton.edu/outages/view.plx?ID=3998

This outage includes all OIT networks including the wireless network in the CS building.

The CS network will effectively be disconnected from the rest of the campus and the Internet during this time. However, if you are on the CS network (e.g., on a wired connection in the CS building), you will be able to reach other systems on the CS Network (even if they are at the remote data center).

During the week of January 28, 2013, we will be moving our centralized computing infrastructure and our cluster (\”ionic\”) to a new location.

For all infrastructure services (e-mail, CS web sites, file services), the downtime window is:

• START: Tuesday, January 29, 2013, at 6:00 AM EST
• END: Wednesday, January 30, 2013, at 10:00 AM EST

If you use the ionic cluster, its downtime window is longer:

• START: Monday, January 28, 2013, 4:00 PM EST
• END: Thursday, January 31, 2013, 10:00 AM EST

Additional Details

\”Infrastructure\” is everything except the ionic cluster and includes e-mail, web pages, file system, printing, and general purpose computing (i.e., penguins: tux and opus; cycles: soak, wash, rinse, and spin).

We are prioritizing the infrastructure over the ionic cluster. We will bring up the ionic cluster within 1 business day of bringing up the infrastructure.

The wired and wireless network in both the CS Building and in the CS section of the data center (e.g., PlanetLab, VICCI, SNS, Memex, etc.) will continue to work during the downtime. Users will be able to access University systems and the Internet from their desktops/laptops during the downtime.

Because the CS e-mail server will be down longer than 4 hours, people sending e-mail to CS accounts will get bounce messages. (These messages are generated by the sending server and are sent back to the sending account.) Properly configured senders will retry sending for 5 days so incoming messages will be delivered after the infrastructure is back online. People sending e-mail to the CS department can expect to see bounce messages of the form \”warning: message not delivered after 4 hours; will re-try for 5 days.\” (The exact message, timeouts, and retry periods are specific to the server sending the message.)

Due to the magnitude of the move, support services from CS Staff will be limited.

While all changes to infrastructure (including this move) have inherent risks, CS Staff has been taking significant steps to reduce these risks to stay within the 28-hour window for the infrastructure and within the additional 1-business-day window for the ionic cluster.

UPDATE 1/28/2013 at 4:08pm: The ionic cluster has been shut down in preparation for its move.

UPDATE 1/29/2013 at 7:06am: All servers have been powered down.

UPDATE 1/29/2013 at 8:47am: All infrastructure servers have been removed from their racks. Packing has begun.

UPDATE 1/29/2013 at 10:25am: Truck with the infrastructure servers has arrived at the new data center; unloading has begun. In the CS Building, the ionic cluster has been removed from its racks; packing has begun.

UPDATE 1/29/2013 at 11:15am: Truck with systems for the ionic cluster has left the CS Building.

UPDATE 1/29/2013 at 11:30am: Installers dropped one of our disk arrays. Assessing damage. Other work continues. For this kind of eventuality, we did engage the manufacturer our storage system vendor and had an on-site engineer already in place to help.

UPDATE 1/29/2013 at 12:42pm: We are working with the vendor to attempt to get a replacement disk array chassis today. Racking of other infrastructure systems continues. Truck unloading of ionic cluster continues.

UPDATE 1/29/2013 at 3:00pm: All systems are in the machine room. Approximately half are in racks. Some are starting to be wired up. Still waiting for arrival of replacement disk array chassis.

UPDATE 1/29/2013 at 4:20pm: All systems are mounted in their racks. Infrastructure systems have been cabled. Awaiting arrival of replacement disk array chassis.

UPDATE 1/29/2013 at 6:10pm: Replacement disk array chassis due to arrive by 8:15pm. We have tested several systems successfully. The disk array chassis is 1 of 7 chassis in our file server system. Even with this setback, we still believe we will meet our deadlines to be back online.

UPDATE 1/29/2013 at 8:25pm: Replacement disk array arrived. Work continues.

UPDATE 1/29/2013 at 9:50pm: Components have been moved from damaged chassis to new chassis. We are working with vendor to bring replacement chassis online.

UPDATE 1/29/2013 at 11:20pm: Chassis replacement complete and system operational. Work continues.

UPDATE 1/30/2013 at 12:10am: Infrastructure services (e-mail, CS web sites, file services) are starting to come back online.

UPDATE 1/30/2013 at 12:40am: All Infrastructure services (e-mail, CS web sites, file services) are now online. The ionic cluster is the only service that is still down. We will be bringing the ionic cluster back online sometime after 1:00pm today.

UPDATE 1/30/2013 at 8:15am: While not all nodes are yet online, the ionic cluster is operational and available for use. The remainder of the nodes that are down will be coming up this morningwhen an additional power strip is installed in one of the cluster racks.

UPDATE 1/30/2013 at 10:55am: The ionic cluster is fully online. At this point, all systems should be operating normally.

One of our database servers is having trouble. This is affecting some of our backend/administrative websites. We are investigating.

UPDATE 3:40pm: problem seems to be a failed (local) hard disk that is part of a mirrored pair.

UPDATE 4/20, 9:00am: due to the hardware failure, the database has become unusable. We are going to re-deploy a replacement server using the last good nightly back-up. During this restore, the following systems will be down:

• The problem system – for today (4/20), send problem tickets to scott@cs rather than csstaff@cs
• Dropbox – previously submitted assignments are not stored in the database and are not affected
• CAS – signin for websites using CS authentication
• CS Guide
• Adm Site – the site where the admin staff updates people information
• Jobs
• Database driven portions of the CS website
• a few other less commonly used pages

Once the databases are back up, we will be auditing them to see what updates may have been made in the interim and taking appropriate action.

UPDATE 4/20, 12:00pm: work continues. We will be restoring the databases to the original hardware (with replaced hard drives). This will get things back to the Wednesday night snapshot. We will start with the restore of a test database. When that is successful, we will be able to start providing estimates for the remainder of the databases and systems.

UPDATE 4/20, 3:15pm: multiple attempts to load the data and yield a stable system have not been successful. We have the data in plain-text form so there\’s no concern about loss of data. In addition, the data loads into the database engine. However, at this time there are lingering issues at the system level (hardware / operating systems). We will provide an additional update by 4pm.

UPDATE 4/20, 4:15pm: attempts to restore with the old hardware, the old operating system, and the old version of MySQL (5.0.x) were not fruitful. We are now installing on new hardware, with a new (puppet-ized) operating system, and a newer version of MySQL (5.1.x). There are slight changes to the meta-data between MySQL versions that require a bit of hand-editing of a couple of files. We are first focusing on the databases needed for dropbox.

UPDATE 4/20, 4:25pm: While our ticket system is still offline, you may use csstaff@cs again to report (other) issues. The e-mail will reach members of CS Staff.

UPDATE 4/20, 5:30pm: Dropbox is back online. Work continues on the databases for other systems.

UPDATE 4/20, 6:00pm: Many systems are back online including the main web site. Work continues on remaining systems.

UPDATE 4/20, 6:20pm: All systems should be operating normally now.

On Tuesday, March 20, 2012, we will have a scheduled downtime from 4:00am to 8:00am EDT.

On Thursday, March 22, 2012, we will have a scheduled downtime from 4:00am to 8:00am EDT.

This downtime affects all users of the department\’s computing and networking infrastructure.

Scheduled work includes:

• Deploy new hardware for cycles (soak, wash, rinse, spin)
• Transfer our virtual machines to new infrastructure
• Upgrade the operating systems on penguins (tux and opus) and cycles to version 6.2
• Deploy a new cluster, named ionic, using hardware from the current c2 cluster (and some of the retired equipment noted above).

SPECIAL NOTE: Since we will be re-installing the operating systems on the public cycle servers (penguins: tux, opus; cycles: soak, wash, rinse, spin), crontab entries will be lost. If you have cron jobs that need to persist, please save them before the downtime and restore them afterwards.

After the downtime, c2 will have approximately 50 servers and ionic will have approximately 20 servers. Over the coming weeks, we will reassign approximately 10 servers per week from c2 to ionic until c2 is empty and then retired. More details will be posted on the CS beowulf list.

This work improves the speed and capacity of our general purpose cycle servers as well as our virtual machine infrastructure. This also synchronizes the operating system version of all our public machines (cycles, penguins, and ionic). The new machines (physical or virtual) will be under the configuration control of our recently deployed Puppet infrastructure. This will allow us to more easily keep the operating environments of the servers both up-to-date and in sync.

During some of the 4am-8am window, most of the services (e.g., e-mail, web, cycle servers) will be unavailable. E-mail destined to the department will be queued and then delivered at the end of the maintenance window.

We have a report of e-mail trouble. We are investigating.

UPDATE 10:05am: E-mail is \”up\” but running extremely slowly leading to time-out problems. We are continuing to work the issue.

UPDATE 11:08am: We are repairing the e-mail system\’s database. The system will continue to be slow for a while. The problem appears to be a latent fault introduced when our previous storage system failed on January 12, 2012. That hardware has since been replaced, but a flaw in the data had migrated to the new system.

UPDATE 11:48 AM: E-mail service is actually down at the moment, while data is being repaired. We are working to restore service ASAP.

UPDATE 14:00: E-mail service is running again, but still very slow. It may be turned off again at some point, as troubleshooting is still underway.

UPDATE 16:00: E-mail service has been restored. We have located a source of delays in our storage and eliminated it for the time being. Our apologies for the extended inconvenience.

On Tuesday, January 24, 2012, we will have a scheduled downtime from 4:00am to 8:00am EST.

This downtime affects all users of the department\’s computing and networking infrastructure.

Scheduled work includes:

• Operating System updates and miscellaneous work on the infrastructure.

As of 7:00am, E-mail and the main web pages are down. We are investigating.

UPDATE 8:05am: Problem is isolated to an old storage server that is used by our virtual machine servers.

UPDATE 8:35am: Storage server is coming back online. After that, all the virtual machines will need a clean restart.

UPDATE 8:45am: Storage server is still having trouble. No ETA yet.

UPDATE 9:35am: We are waiting for a return call from the storage server vendor.

UPDATE 10:05am: Systems are beginning to come back online. Simultaneously, the storage system is rebuilding and remirroring the data. During this time, we expect that systems will be slower. We are also working to stabilize the systems.

UPDATE 11:20am: Systems are basically online. Storage system rebuilding (and associated slowness) continues. E-mail was delayed but none lost.

On Tuesday, December 20, 2011, we will have a scheduled downtime from 4:00am to 8:00am EST.

This downtime affects all users of the department\’s computing and networking infrastructure.

Scheduled work includes:

• Moving several infrastructure file systems to our new storage system. This requires a configuration change to most of our servers that can only be done when the systems are idle.