Open Infrastructure

How do we plan to handle failures in etcd¶

Failures can be temporarily (leadership change) or permanent (etcd cluster dies)

We use four type of functions of etcd
1. get
2. put
3. get_prefix
4. watch_prefix

get¶

• EtcdEntry
• Exception (KeyError if key not found, and a pleothera of other depending on situations)

TODO: KeyError is recently introduced so would require us to handle it in all stretches of uncloud other errors are also not handled.
TODO: Handle Exception thrown by get.

put¶

• Response from etcd library
• Exception

TODO: See whether the response from etcd library has success/failure indication.
TODO: Handle exception in case something goes wrong.

get_prefix¶

can output/yield the following:

• EtcdEntry
• Exception

TODO: Handle exception in case something goes wrong.

watch_prefix¶

can output/yield the following:

• EtcdEntry
• an empty iter in case of exception

I am little uncertain how to handle failures in etcd. For Example, put every etcd function call in try/except block or do something else.