Feature #3944: Add Fiber#root? method

• Status: Rejected
• Assignee: ko1 (Koichi Sasada)
• Target version: -

[mperham (Mike Perham)]

Since the root fiber is treated differently than other Fibers (e.g. you can't yield from the root), code which can optionally use fibers for high-performance IO (in my case, the Mysql2 driver) needs Fiber#root? to know when it can use Fibers to execute a query versus using a standard blocking call.

[naruse (Yui NARUSE)]

• Status changed from Open to Assigned
• Assignee set to ko1 (Koichi Sasada)

[godfat (Lin Jen-Shin)]

Any progress on this?

[ko1 (Koichi Sasada)]

• Description updated (diff)

Sorry for long absent.

I'm not sure why Fiber#root? is needed.
Could you give the examples?
I think if you don't use Fiber#root, then Fiber#root? is not needed.
And I can't understand why Fiber#root is needed.
(I think Fiber#root is for Fiber#transfer)

[ko1 (Koichi Sasada)]

• Status changed from Assigned to Feedback

[ko1 (Koichi Sasada)]

• Status changed from Feedback to Assigned
• Target version set to 3.0

[schmurfy (Julien A)]

what is Next Major ? 3.0 ???
The issue was more than 2 years ago, after looking at the code the implementation should not be a challenge and yet nothing.
it is so depressing watching the ruby redmine seriously...

[zzak (zzak _)]

Julien please see ruby-core:45849 and also read up on the wiki: http://bugs.ruby-lang.org/projects/ruby/wiki/HowToRequestFeatures

[ko1 (Koichi Sasada)]

• Status changed from Assigned to Feedback

[jjyr (Jinyang Jiang)]

Fiber#root is useful.
For example, I want write a method, when under EM::Synchrony environment it should use EM::Synchrony call(in fact is async call, and writing in sync, power by fiber), and when under Fiber#root, it should use sync call, so I need detect whether current fiber is root fiber.

[ko1 (Koichi Sasada)]

(2014/05/23 15:17), [e-mail] wrote:

For example, I want write a method, when under EM::Synchrony environment it should use EM::Synchrony call(in fact is async call, and writing in sync, power by fiber), and when under Fiber#root, it should use sync call, so I need detect whether current fiber is root fiber.

Could you break down the description?
I don't know EM::Synchrony.

Why you need to choose root or not root?

--
// SASADA Koichi at atdot dot net

[Nakilon (Victor Maslov)]

For example, the Facebook Graph API server accepts POST request to calculate smth heavy and returns id of the 'async job'. Then I GET status until it's "complete" and then I GET the actual result.
To process thousands of requests I call my method_that_gets_data from inside Fiber.new for each URL and cycle through them until they are done. Inside method_that_gets_data there are yeilds when doing the first POST and when status is not "complete", so I know when the fiber is done or I should just check another one.
Now I want to reuse the method_that_gets_data without fibers -- to know whether to yield or to work until done I need to know if I'm running inside Fiber or not.

[nobu (Nobuyoshi Nakada)]

• Description updated (diff)

Couldn't anyone explain how the method will be used and why it will be necessary?
As far as I read, it sounds that it's EM specific and EM should take care of it.

[spatulasnout (B Kelly)]

Nobuyoshi Nakada wrote:

Couldn't anyone explain how the method will be used and why it will be necessary?
As far as I read, it sounds that it's EM specific and EM should take care of it.

It might indeed be EM-specific.

I have an RPC library based around EventMachine and fibers. It needs to record
Fiber.current at the point EventMachine.run is called.

The library then uses RPC.on_em_fiber? (i.e. Fiber.current == @em_fiber)
to make necessary choices similar to those described by earlier posters.

If this is indeed equivalent to what earlier posters have requested, it does
seem like functionality that could be provided by EM or EM::Synchrony.

Regards,

Bill

[Nakilon (Victor Maslov)]

It might indeed be EM-specific.

I didn't use EM and not going to in the case I've described above.

[christopheraue (Christopher Aue)]

I also need Fiber#root?. My Scenario looks similar to the ones above, but I'm not using EM:

I have a loop watching sockets. This loop runs in a fiber. Each time a socket is readable another fiber is created to handle the request.

So, if I have a server only listening for requests:

1. I resume (i.e. start) the loop fiber,
2. resume (i.e. start) a dedicated fiber for each request,
3. yield to the loop fiber when the request itself needs to send a request to another server and must wait for the response,
4. resume the request fiber again once the response is there,
5. go to step 3) until the request is fully handled
6. yield to the main loop and go to step 1)

The loops starts at the beginning and runs forever. But that is what I expect a server to do. No problem.

But there is a second scenario, when I have a linear script with start and finish. This script uses the same code as the server above and during its course calls the same methods the server uses to handle its requests. Calling these methods sends the same requests to another server for which responses the script needs to wait. After the request is fully processed the script continues. Then the flow looks like this:

1. I'm processing the script on the root fiber.
2. When a request in need of a response from a server is encountered, I'm actually at a point corresponding to step 3) from above. But, I cannot yield back to the loop since it has not been started yet. I have to resume it.
3. From here, I'm at step 1) from above and can proceed accordingly.
4. Once my request is fully handled I need to yield from the loop to the root fiber to continue the script.

To decide if I need to resume the loop or yield to it, I need to know if the request originated from the root fiber or from one of the request fibers of the loop.

For a solution now, I have created a small gem monkey patching Thread and Fiber to get the root fiber: https://github.com/christopheraue/ruby-thread_root_fiber

[ko1 (Koichi Sasada)]

Sorry for very late response.

Victor Maslov wrote:

Now I want to reuse the method_that_gets_data without fibers -- to know whether to yield or to work until done I need to know if I'm running inside Fiber or not.

You shouldn't rely on root? because some libraries (for example, some kind of Enumerator objects) use fibers.

[ko1 (Koichi Sasada)]

• Status changed from Feedback to Rejected

To decide if I need to resume the loop or yield to it, I need to know if the request originated from the root fiber or from one of the request fibers of the loop.

For same reason of #16, you shouldn't use it.