Kubernetes logs are not strictly sequential

[timothyb89]

Logs from the Kubernetes reader aren't strictly sequential. This makes output potentially confusing, but isn't really trivial to solve.

Log messages from different pods obviously don't need to arrive at the same time since we're grabbing all logs since a pod was started, in parallel. We can't change stdout renderers' output after a message is printed, plus it'd be challenging to inform users that a new message arrived some arbitrary number of pages into the past even if we could.

The only 'easy' solutions I can immediately think of:

• hold messages in a buffer for a bit and try to ensure output is ordered
  • we'll still have to release out-of-order messages eventually (unless discarding is an option...) but at least we can improve the situation
  • possibly 10s buffer at first run and reduce it once we've been running, to allow for initial out-of-order log ingest. logs tend to be ordered once we've caught up on the backlog
• in addition to a buffer, add an option to discard or somehow flag messages as being out-of-order
• alternatively / additionally, support random insertion in the interactive viewer. this would most likely seriously hurt performance, and have bad UX

A --tail option will mitigate this issue a bit (#33)