ktoso
commented
7 years ago 
This ties in with the proposals in (17694) https://github.com/akka/akka-http/issues/105 and should probably approached "holistically".
In general, several parsing modes are possible that would determine what kind of entities the user can expect when receiving an HttpMessage:
- the current behavior, with entity types very closely fitting what comes in on the wire
- one disadvantage is the somewhat unnatural and non-deterministic distinction between Strict and Default
- each chunk is currently aggregated before it is dispatched
- only Strict messages (this ticket, I like the proposed configuration options)
- only streamed messages (either with or without content-length available), see (17694) https://github.com/akka/akka-http/issues/105
- Chunked entities would be auto-converted to Default-ones to prevent having to collect full chunks
stefanasseg
yeryomenkom
Currently the core layer only strictifies incoming message entities if the response bytes are fully contained in the chunk of bytes that have currently been read from the receive buffer. Since this is non-deterministic the user always has the responsibility for "draining" the entity bytes from the connection in order to free it up for subsequent messages.
The proposal is to introduce a new config setting called
akka.http.parsing.auto-to-strictwhich increases user convenience for the standard case (when the entity is "small") and at the same time removes all non-determinism.Possible values for the setting:
offThe parser never creates strict entities, even if the bytes are "already there".onThe parser always strictifies unchunked entities. Boundedness is achieved solely via themax-content-lengthsetting.<int value>The content-length threshold up to which the parser automatically strictifies the entity.