On the first job invoked by a sidekiq process, the librato-sidekiq middleware will drop (cause sidekiq not to execute) the middleware immediately after it in the chain. Here's why it happens:

When a job is ready to be invoked, Sidekiq.server_middleware.invoke is called with the relevant worker instance and arguments. #invoke maps over the middleware chain and instantiates new middleware objects of the correct classes (dups the chain, essentially).

Each time the librato-sidekiq middleware is initialized it "reconfigures" itself, which really means that it removes and re-adds itself to the middleware chain entries array. Array#each expects that the position of objects in the array will not change during iteration. Since the librato-sidekiq middleware removes an element of the entries array while the chain is mapping over that array, the entry immediately following ends up "skipped" (and the librato-sidekiq middleware runs twice).

Is there a reason why #reconfigure has to be run every time the middleware is instantiated? Can we move the method to the class and just run it once after .configure is run?

The offending code is here: https://github.com/StatusPage/librato-sidekiq/blob/master/lib/librato-sidekiq/middleware.rb#L46-54

For reference, here is the sidekiq class involved: https://github.com/mperham/sidekiq/blob/master/lib/sidekiq/middleware/chain.rb

And here is the logic for Array#each: http://www.ruby-doc.org/core-2.1.2/Array.html#method-i-each