On-line detection of a part of a sequence with unspecified distribution

We register a Markov process. At two random moments θ1θ1, θ2θ2, where θ1<θ2θ1<θ2, the distribution of the observed sequence changes. It is known before θ1θ1 and after θ2θ2, but between these instants is unknown, chosen randomly from a set of distributions. The optimal stopping rule which stops observation of the sequence between disorders θ1θ1 and θ2θ2 is identified.