FEATURE REQUEST: Timed Software Upgrades

[mhoppes]

I don't know if this is even feasible with the current software, but it would be nice to be able to load a firmware image into the system and program it to update at a certain time (or after a countdown if a time is too hard/dangerous).

This would allow me to sleep through the night and still update my switches :)

[sirhc]

Yea it is possible.
However the switch ACTUAL down time from an upgrade is like 17 seconds?

On Sunday before Eric caught his plane back to Canada we were eating lunch discussing ways to allow 0 seconds down town on a firmware upgrade unless there was a Kernel, driver or API upgrade in the firmware upgrade.

The argument is rather it is worth the time to develop it?
17 seconds is pretty quick and we can reduce that to about 10 to 14 seconds which we probably will before v1.0 is released.
All the time the switch unpacking and writing to flash the core remains up and passes traffic.

Also once the file is uploaded the browser can be closed as the upgrade is NOT dependent on the browser that initiated the upgrade because the SOC takes over the upgrade process once the file uploads, I requested this little feature!

[mhoppes]

To this point, I'll add I had noticed this. The upgrade process is EXTREMELY quick and down time was a hand full of pings. KUDOS to whomever managed this? Eric? Dave?

The thought process for this is not that it's 10 seconds of down time, but rather if something goes wrong you want it to be at 3am, not 10am. Carrier grade networks just don't do upgrades in the middle of the day, no matter how small the downtime is.

I'll also add that bootup time on these switches is amazingly fast as well! Very cool.

[mhoppes]

Since you aren't actually writing anything to the switch core... is it possible to just reboot the GUI/Kernel CPU without touching the switch chip? That would allow for 0 down time upgrades, no?

[lligetfa]

This would allow me to sleep through the night and still update my switches :)

I would lay awake wondering if something borked during or after the upgrade. I just find that there are too many dependencies and that sometimes kit needs to be rebooted in a particular sequence.

On my last job I worked the late shift applying patches to hundreds of servers. It almost always resulted in some seat-of-the-pants flying. One time I knocked off early thinking things would complete and walked in to a surprise in the morning. I had a script running to delete snapshots but some of the snapshots grew too fast and ran out of storage.

[sirhc]

Since you aren't actually writing anything to the switch core... is it possible to just reboot the GUI/Kernel CPU without touching the switch chip? That would allow for 0 down time upgrades, no?

Depends on what you have the switch doing.
There are routines running on the SOC that control things like Aggregation, STP, Loop Protection, and such that are controlled outside the core that would have to be addressed so that the core could work on it's own for 17 seconds. Not impossible but a lot of work as I said.

Normal things like VLANs, port configurations and such are handled inside the core yes.

Hey Matt you put this post in the wrong forum, it should be under WISP Switch not TOWER Switch!

[mhoppes]

Hmmm you're right... I'll get the moderator to move it :)

[rebelwireless]

IMHO, this is a recipe for a long outage. I believe that all firmware upgrades should be interactive so that you can react to a failed update right now instead of when the alarm clock buzzes :/