mlococo
commented
9 years ago Thanks for your report. I'm not aware of a way to pass in a list of seed nodes, though I'm recent enough as a maintainer that I might have missed something myself.
Unfortunately, I'm unlikely to develop this feature myself and unlikely even to merge a PR for it. ComboAMI has a number of shortcomings that make it unsuitable for use in long-lived production clusters:
- Not automated test suite, means it's risky to run in prod for change-control reasons. We release fairly naughty bugs sometimes and rely on the community to find them.
- No help for upgrades, which must be done completely by hand.
In the context of the flaws above, seed management for new nodes seems like a pretty minor issue. I'm inclined to keep long-term management oriented features out of ComboAMI.
As a workaround, the --repository feature does let you fork the comboAMI repository, which is git-pulled on boot. So you could fork and hack in whatever seed handling you want and then pass --repository https://github.com/your-repo#yourbranch to boot your customized version.
Closing this wontfix
Hi,
I am using DS AMI (2.5) as base image for a cluster. I am trying to configure EC2 user data in such way that I can change the capacity of the cluster just by launching more instances with the same user data and no manual intervention. I am stuck right now because of the reflector (it overwrites the list of seed nodes). Perhaps I overlooked something, is there a way to provide the list of seed nodes on user data and skip the reflector stage altogether?