[
https://issues.apache.org/jira/browse/THRIFT-4405?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16754021#comment-16754021
]
James E. King III commented on THRIFT-4405:
-------------------------------------------
It turns out the languages / combinations above do not need sequence IDs for
proper operation. [~codesf] and [~jensg] I think we should refine the
requirements here as follows, let me know if you agree or we should discuss
more:
# A client *{{MUST}}* send a sequence ID with each request (this is part of the
protocol header, although I am not sure about text-based ones).
# A client *{{MAY}}* make each sequence ID unique, if it requires sequencing
for proper operation. This is typical for clients that allow for multiple
outstanding requests. Clients that do not require sequencing are
*{{RECOMMENDED}}* to send zero for a sequence.
# A server *{{MUST}}* provide the same sequence ID with any response related to
that request.
# A server *{{MUST NOT}}* assume all sequence IDs will be unique.
Thoughts?
> Proper use of sequence IDs in all clients and servers
> -----------------------------------------------------
>
> Key: THRIFT-4405
> URL: https://issues.apache.org/jira/browse/THRIFT-4405
> Project: Thrift
> Issue Type: Improvement
> Components: Test Suite
> Affects Versions: 0.10.0
> Environment: docker ubuntu-xenial
> Reporter: James E. King III
> Assignee: James E. King III
> Priority: Major
>
> Create a feature test that verifies sequence numbers are used properly.
> Write a server that verifies clients are generating unique sequence IDs.
> Write a client that makes sure servers return the same sequence ID that was
> given.
> To do this, I enhanced the C++ TProcessorEventHandler class to include a
> preReadSeq, which is like preRead but carries the sequence ID.
> In the C++ TestServer, I check to see if the sequence numbers are unique and
> do not repeat; if any of them do, the cpp test fails.
> The following languages properly send sequence IDs (for the binary protocol):
> * dart
> * go
> * nodejs
> * java
> * rs
> The rest of the languages do not. Now, one could argue that unless a
> language has a concurrent-safe client and server, sequence IDs are
> unnecessary. While that is true, all languages should respect that the
> protocol has a sequence ID and there could be future implementations that
> will require all clients are well-behaved, which is why I am putting this
> test in.
> Languages fixed up so unique sequence IDs are sent by the client, and
> verified by the tests:
> * cpp (was only sending unique sequence IDs for Concurrent clients, now it
> does for the regular one too)
> * csharp (seqid_ was not bring incremented with each use)
> * lua (seqid_ was not bring incremented with each use)
> * perl (seqid_ was not bring incremented with each use)
> * ruby (seqid_ was not bring incremented with each use and a unit test was
> updated to no longer be pending)
> Languages left to do:
> * c_glib
> * erlang
> * haskell
> * php
> * python
> * python3
> * any non-cross tested languages
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
