Zuul2 integration

[wavesZh]

Describe what this PR does / why we need it

integration of zuul2.

Resolves #80

Describe how you did it

As Zuul 2.x is based on netty, a event-drive model, so we use AsyncEntry to do flow control.

• SentinelZuulInboundFilter: This inbound filter will regard all proxy ID (proxy in SessionContext) and all customized API as resources. When a BlockException caught, the filter will set endpoint to find a fallback to execute.
• SentinelZuulOutboundFilter: When the response has no exception caught, the post filter will trace the exception and complete the entries.
• SentinelZuulEndpoint: When an exception is caught, the filter will find a fallback to execute.

Describe how to verify it

run test and demo

Special notes for reviews

[sczyh30]

👍

[codecov-io]

Codecov Report

Merging #1138 into master will decrease coverage by 0.57%.
The diff coverage is 10.73%.

@@             Coverage Diff              @@
##             master    #1138      +/-   ##
============================================
- Coverage     43.01%   42.44%   -0.58%     
- Complexity     1567     1581      +14     
============================================
  Files           337      353      +16     
  Lines          9877    10082     +205     
  Branches       1332     1352      +20     
============================================
+ Hits           4249     4279      +30     
- Misses         5099     5275     +176     
+ Partials        529      528       -1

Impacted Files	Coverage Δ	Complexity Δ
...ateway/zuul2/api/route/PrefixRoutePathMatcher.java	0% <0%> (ø)	0 <0> (?)
...ter/gateway/zuul2/api/route/ZuulRouteMatchers.java	0% <0%> (ø)	0 <0> (?)
...uul2/api/matcher/HttpRequestMessageApiMatcher.java	0% <0%> (ø)	0 <0> (?)
...nel/adapter/gateway/zuul2/filters/EntryHolder.java	0% <0%> (ø)	0 <0> (?)
...ul2/filters/inbound/SentinelZuulInboundFilter.java	0% <0%> (ø)	0 <0> (?)
...way/zuul2/callback/DefaultRequestOriginParser.java	0% <0%> (ø)	0 <0> (?)
...way/zuul2/callback/ZuulGatewayCallbackManager.java	0% <0%> (ø)	0 <0> (?)
...gateway/zuul2/api/route/RegexRoutePathMatcher.java	0% <0%> (ø)	0 <0> (?)
...y/zuul2/filters/endpoint/SentinelZuulEndpoint.java	0% <0%> (ø)	0 <0> (?)
...er/gateway/zuul2/HttpRequestMessageItemParser.java	0% <0%> (ø)	0 <0> (?)
... and 25 more

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 49861f4...162331d. Read the comment docs.

[locuslin]

I'm working on Zuul2 intergration too.I'm doubt that there are problems.
1.Importing "spring-core" just to use the AntPathMatcher is a big cost .Might use ant-style-path-matcher) Instead?
2.SentinelZuulOutboundFilter may run in a different thead from the inbond one,is that OK just calling "ContextUtil.enter"&"Context.exit"?I read the code in sentinel-spring-cloud-gateway-adapter,but have't find a solution.

[wavesZh]

I'm working on Zuul2 intergration too.I'm doubt that there are problems.
1.Importing "spring-core" just to use the AntPathMatcher is a big cost .Might use ant-style-path-matcher) Instead?
2.SentinelZuulOutboundFilter may run in a different thead from the inbond one,is that OK just calling "ContextUtil.enter"&"Context.exit"?I read the code in sentinel-spring-cloud-gateway-adapter,but have't find a solution.

1. That is better to use a lighter AntPathMatcher, but i need more suggestions.
2. Because of event-drived model, one thread may be corresponds to many sentinel's contexts, so ThreadLocal is invalid to deliver context for a request.ContextUtil#enter create a ThreadLocal context as a head of invaction chain. The context has store in AsyncEntry when do SphU.asyncEntry and I use zuul's context to deliver and use it. ContextUtil#exit just clean ThreadLocal. but in my case, ContextUtil#exitmay not need to be invoked inSentinelZuulOutboundFilter`.

[locuslin]

It's better to call "setErrorEndpoint" and "setShouldSendErrorResponse" here.

[wavesZh]

If you set "setShouldSendErrorResponse" with true, the rest of inbound filter will not be invoked. I want that user can control the invocation of inbound filter by ZuulConstant.ZUUL_CTX_SENTINEL_BLOCKED_FLAG. It can be more flexible.

[locuslin]

1.Generally when a resource is blocked,the invocation will be aborted.I think it would be confusing if the invocation continues.It seems better to make this feature enabled by configuration and disabled by default？
2. I finally work out the context switching(not tested although) , may you have a look at my code?I think it may case a memory leak here,and I tried to fix it.

3. By the way，the code format is different from the orgin code ,especially for the indents.

[wavesZh]

1. Your suggestion is well, i will optimize it.
2. I see a little your code, what is meaning about the code. asynchronous action should be done in asynchronous context to keep a correct invocation chain.

[locuslin]

Just refer to the SentinelReactorSubscriber,exactly hookOnSubscribe and hookOnNext.It's not necessary to do like this,but just in case and to keep the same style.This is the document.

[locuslin]

need to deliver the context to "SessionContext" and exit the context in outbound filter.I doubt "ContextUtil.exit" can't exit actually because that the "AysncEntry" haven't exit,I will confirm it later. And there is a limit in ContextUtil,and there will be a memory leak if not clean correctly.

[wavesZh]

threadlocal of context may will not clean up in your case, i will fix it. And i think the place of ContextUtil.exit() is not important in event-driven model, it's function is just to clean threadlocal, but it maybe necessary to store the context for other purpose instead of discarding it.

[locuslin]

Also a suggestion ,it seems better only operate the context in Entry and don't store the context in ThreadLocal ,just as dubbo did.I found it hard to use the context in asynchronous invocation .But it may be a big change 😔.

[sczyh30]

Could you please resolve the conflicts?

[sczyh30]

Is this the standard and universal key for routes of Zuul 2.x?

[locuslin]

That's not.The standard method is to call com.netflix.zuul.context.SessionContext#getRouteHost，and the key is "routeVIP",so there is a bug here.

[sczyh30]

1.Importing "spring-core" just to use the AntPathMatcher is a big cost .Might use ant-style-path-matcher) Instead?

If there are any popular lightweight AntPathMatcher implementations (with correct functionality and good performance), we could replace spring-core with it.

[locuslin]

1.Importing "spring-core" just to use the AntPathMatcher is a big cost .Might use ant-style-path-matcher) Instead?

If there are any popular lightweight AntPathMatcher implementations (with correct functionality and good performance), we could replace spring-core with it.

The most popular and suitable implementation I could find is the ant-style-path-matcher.There are some similar test cases (see here)borrowed from spring,you can refer to it and decied whether to use it.

[sczyh30]

LGTM. I'll improve some of the code later.

[sczyh30]

Fabulous work. Thanks for contributing!