BftSmart not ordering well when too many requests come in. about library

[jcs47]

What do you mean by being out of order? Are you saying that each replica receives messages by different order, or that FIFO order in respect to each client is not guaranteed? Please be more precise.

[Raycoms]

I'm currently debugging this, is it possible that appExecuteBatch gets called concurrently for each batch? 2018-03-30 19:36 GMT+01:00 jcs47 <[email protected]>:

…

What do you mean by being out of order? Are you saying that each replica receives messages by different order, or that FIFO order in respect to each client is not guaranteed? Please be more precise. — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub <#56 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AGI9y5vQGsJJByDKj14MB488pI-gEAE0ks5tjnsUgaJpZM4TB_Pw> .

[jcs47]

No

[Raycoms]

So the next batch only gets called after I returned the response for the current batch, right?

[jcs47]

Exactly. But bear in mind, you can get batches of different sizes across different replicas. However, the order by which the messages are ordered should be the same.

[Raycoms]

I'm currently debugging this, and it's incredibly strange, All replicas receive the same data until a certain moment, then, out of nothing, 1 replica starts getting completely different data than the other 3

[jcs47]

Which replica? Maybe what you are seeing is a replica getting late in relation to all others. This is a common phenomenom. Try to halt all clients and allow all replica to finish processing.

[Raycoms]

It is each time another replica, it seems like all messages arrive in order, but 1 of the messages isn't getting executed for some reason.

[jcs47]

Like I said, check if what you are seeing is a delayed replica.

[Raycoms]

I found it... Somewhere the code was going down, for some reason I didn't get an exception again and this stopped the execution of the next message in the batch.