wmantly / lxc_manager_node Goto Github PK

user_data String A string containing 'user data' which may be used to configure the Droplet on first boot, often a 'cloud-config' file or Bash script. It must be plain text and may not exceed 64 KiB in size.

[12:48 PM]
that can be added to the API call to create the droplet

[12:48 PM]
Thomas Harvey
ok

[12:49 PM]
William Mantly so, you can set up a route on the manager (edited)

[12:49 PM]
and it can "call home" when its done

[12:49 PM]
Thomas Harvey ok

[12:50 PM]
William Mantly
hell, that can even do away with the isactive polling

[12:50 PM]
Thomas Harvey that sounds good. how long should the manager wait before killing ignorant offspring?

[12:50 PM]
William Mantly
10mins

[12:50 PM]
maybe

[12:51 PM]
Thomas Harvey
should that be a max time without checking in?

[12:51 PM]
William Mantly hmm

[12:51 PM]
not sure

[12:52 PM]
Thomas Harvey like killing a worker who has not checked in in the last 10 to 15min?

[12:52 PM]
William Mantly
yea

[12:52 PM]
sounds good

[12:53 PM]
Thomas Harvey
is this going to increase or decrease bandwidth usage? im thinking these additional calls actually reduce the amount of overall chatter.

[12:59 PM]
Thomas Harvey i know this is a silly concern, but if we have the runners send the data we will need them to have unique tokens as well. like auth tokens

[1:00 PM]
so that will need to be added to the boot script

[1:02 PM]
William Mantly yea

[1:02 PM]
100%

[1:02 PM]
and the over chatter will be less

[1:02 PM]
less calls on the manager too

[1:02 PM]
Thomas Harvey and the manager will still need to check in with the workers to sync runner counts

[1:03 PM]
and that will need to be more often than 10 min

[1:03 PM]
William Mantly maybe the workers should check in

[1:03 PM]
Thomas Harvey yeah thats what i was about to say

[1:04 PM]
maybe the workers take all that responsibility and the manager just tells the worker when to free runners

[1:05 PM]
so 10 min max dark time and have the cronjob report back a list of runners every time it runs.

[1:06 PM]
manager keeps track of last communication timestamp

[1:07 PM]
this is kind of a major change. should i just get it working with maybe a simple boot script that tells us its booted. then after that works we can move more responsibility to the worker. thoughs?

[1:09 PM]
the changes we are suggesting will touch a lot of code while a simple boot script that tells us when the boot is complete will touch much less code.

[1:10 PM]
William Mantly
yea

[1:10 PM]
1 thing at a time

[1:11 PM]
Thomas Harvey
these should be issues

[1:12 PM]
William Mantly yes

new messages
[1:12 PM]
i agree

Refactor scaling logic out of workers object.

• makes creating and implementing new scaling strategies easier
• allows the strategy to change dynamically