That's quite an interesting architecture. Do you have data on how frequently your spot instances terminate gracefully (with time to finish their requests) vers

That's quite an interesting architecture. Do you have data on how frequently your spot instances terminate gracefully (with time to finish their requests) versus how often they zap out?