outputs.graphite: Retry sending metrics immediately after reconnect

[piotr1212]

If writing to Graphite would fail the plugin would reconnect, but not retry to send the metrics until the next interval. In a situation when the connection would break before the next interval the metrics would never reach the Graphite server. Obviously this is a network issue which but Telegraf should handle these kind of situations better.

This patch retries to send the metrics one time immediately after reconnecting.

Required for all PRs:

• Signed CLA.
• Associated README.md updated. <- not needed
• Has appropriate unit tests.

[danielnelson]

Can you check if your disconnects are detected by the code in checkEOF? I think it might make more sense to reconnect before Write, so that we don't need to double send.

[piotr1212]

Yes it does detect them, but I wasn't sure if I have to check all connections and then reconnect the broken ones, reconnect all connections or be happy with at least one connection. I suspect I would have to refactor connect() then as well, as it now reconnects all connections. Or add some way to mark a connection as broken so it won't try to send on it. I thought this would be safe as well, there might be situations which are not detected by checkEOF. I'll think a bit about it, if you have any suggestions let me know.

And this doesn't double send, the second send is only called when the first failed.

[danielnelson]

Okay, I think we will want to refactor this code in the future so that we can reconnect individually to each server. This way the logic can basically be:

for each server
  if not connected
    connect
  write

This is still an improvement, so I'm going to merge.