twitter-archive/kestrel

Reliable fetch question

Closed this issue · 1 comments

The documentation explains how using /open and /close for reliable fetching of items could potentially cause clients to receive duplicates. I'm guessing there's a maximum number of seconds that the Kestrel server waits for between putting an item aside for an /open request and making that item available after not receiving any /close requests for that connection. In other words, how long is the window of time between /open and /close, such that duplicates are not created?

I was wondering what that number is, and if it is in any way configurable?

in the thrift API, a client passes the timeout along with the GET request.

over memcache and the text protocol, the item stays open until the client either confirms, aborts, or disconnects. a disconnect is treated as an abort.