1) The log streams, with timestamps as accurate as is possible, are a complete sequence of traffic packets or requests and responses from a source cluster and its clients. Log Streams can have two different durability guarantees, which are described in subsequent diagrams.
2) We may need to restart replays to new clusters and captures/tests could cover weeks' worth of data for some customers. Directing all of the captured logs to long-term storage allows us to replay from the beginning of the test to a new cluster.
3) Amendments may be added between replay runs. This is where data . These may be in the form of additional messages or transformations (split msg X into new messages Y & Z).
4) Amendments to the replay stream may be made as data is being processed if the traffic comparator determines that a may have yielded different results due to transient/causality issues. In such cases, it may issue a retry to improve the likelihood of a match and/or to give the customer more information.
5) Replayers may be seated in one network or across many. They are fed streams that are segmented by index, with mutations (PUT, UPDATE, DELETE) for each index being handled on no more than one replayer.