Hello,
we started discussing about removing MI (so called management interface) for very long time, more or less since 2008. The RPC should remain the control interface, given its better structure for commands, parameters, etc ... MI is custom protocol using a line-oriented communication via fifo or socket file with kamailio (e.g., implemented mi_fifo and mi_datagram modules). RPC is the alternative, a more standardized concept, with better structured format.
I think it's time to set a clear roadmap for doing the removal. Overall, it will be easier to maintain the code, right now being duplicated code for doing the same operation over MI or RPC, and MI shows its limitations (or complexity to deal with) for advanced needs (see the discussions about how to provide multi-line value parameters over MI).
So, I want to know if there are many relying on MI directly and they still want to keep it, what would be the expected duration they need for upgrading their tools to work with RPC interface, other relevant aspects people have in favour of mi vs rpc.
I am even willing to do the removal in time befire freezing the 5.0 branch. We will ensure a clean start of 5.x series.
The main concern from my point of view is kamctl -- but I think we can preserve the compatibility for kamctl commands and parameters (so command line execution of kamctl will be the same), but the output might be different. That's because it should be easy to updated it to communicate with jsonrpc-s module, but then it will get json-formatted results.
To summarize, two big questions to answer:
a) Are you ok to remove the MI code/commands?
b) If yes to a), are you ok to be done for v5.0?
Not providing feedback will be considered as 'yes' for both questions, so **speak up if you want MI to be kept or delay it removal**.
Cheers, Daniel
+1 - I'm all for cleaning up any technical debt and moving on with more normalized concept.
:)
Sincerely, Brandon Armstead
On Thu, Dec 1, 2016 at 6:17 AM Daniel-Constantin Mierla miconda@gmail.com wrote:
Hello,
we started discussing about removing MI (so called management interface) for very long time, more or less since 2008. The RPC should remain the control interface, given its better structure for commands, parameters, etc ... MI is custom protocol using a line-oriented communication via fifo or socket file with kamailio (e.g., implemented mi_fifo and mi_datagram modules). RPC is the alternative, a more standardized concept, with better structured format.
I think it's time to set a clear roadmap for doing the removal. Overall, it will be easier to maintain the code, right now being duplicated code for doing the same operation over MI or RPC, and MI shows its limitations (or complexity to deal with) for advanced needs (see the discussions about how to provide multi-line value parameters over MI).
So, I want to know if there are many relying on MI directly and they still want to keep it, what would be the expected duration they need for upgrading their tools to work with RPC interface, other relevant aspects people have in favour of mi vs rpc.
I am even willing to do the removal in time befire freezing the 5.0 branch. We will ensure a clean start of 5.x series.
The main concern from my point of view is kamctl -- but I think we can preserve the compatibility for kamctl commands and parameters (so command line execution of kamctl will be the same), but the output might be different. That's because it should be easy to updated it to communicate with jsonrpc-s module, but then it will get json-formatted results.
To summarize, two big questions to answer:
a) Are you ok to remove the MI code/commands?
b) If yes to a), are you ok to be done for v5.0?
Not providing feedback will be considered as 'yes' for both questions, so **speak up if you want MI to be kept or delay it removal**.
Cheers, Daniel
-- Daniel-Constantin Mierla www.twitter.com/miconda -- www.linkedin.com/in/miconda Kamailio World Conference - May 8-10, 2017 - www.kamailioworld.com
SIP Express Router (SER) and Kamailio (OpenSER) - sr-users mailing list sr-users@lists.sip-router.org http://lists.sip-router.org/cgi-bin/mailman/listinfo/sr-users
Brandon Armstead writes:
+1 - I'm all for cleaning up any technical debt and moving on with more normalized concept.
I'm also in favor of getting rid on MI interface. I'm myself relying only on one MI command (pua_publish) for which an RPC command does not exist (or did not exist last time when I checked).
Some years ago I tried to implement RPC version of pua_publish, but was not able to get the call back working correctly. I also sent some messages including a piece of code to the dev list, but it was a dead end then.
Very nice if pua.publish would see a day of light again.
-- Juha
I would also say that at this point, almost all classical uses of MI that I know of can be met with RPC.
If, in removing MI for v5.0, something should be overlooked, I trust any howls of protest will expediently illuminate any small things that fell through the cracks. The major functionality for most modules is clearly there, though.
-- Alex
I would also say that at this point, almost all classical uses of MI that I know of can be met with RPC.
Alex,
Did you see my message about pua_mi? Is there now rpc version of it that I have missed to notice?
-- Juha
On 02/12/2016 00:24, Juha Heinanen wrote:
Brandon Armstead writes:
+1 - I'm all for cleaning up any technical debt and moving on with more normalized concept.
I'm also in favor of getting rid on MI interface. I'm myself relying only on one MI command (pua_publish) for which an RPC command does not exist (or did not exist last time when I checked).
Some years ago I tried to implement RPC version of pua_publish, but was not able to get the call back working correctly. I also sent some messages including a piece of code to the dev list, but it was a dead end then.
Very nice if pua.publish would see a day of light again.
Can you find that discussion in the archive? Open an item on bug tracker and add the link to the discussion there, so we don't forget about it.
Cheers, Daniel
On donderdag 1 december 2016 15:17:18 CET Daniel-Constantin Mierla wrote:
RPC is the alternative, a more standardized concept, with better structured format.
Before removing MI and letting everyone move to RPC, it might be wise to go over all RPC interfaces and fix them to be neat , sane and somewhat consistent. For example, there are still interfaces coding arrays as hashes with multiple identically named fields (htable.dump is one i recently ran into) . There is no language and/or JSON/XML library i know of that can properly handle those (most of them will just end up with the last entry).
On 02/12/2016 11:21, Alex Hermann wrote:
On donderdag 1 december 2016 15:17:18 CET Daniel-Constantin Mierla wrote:
RPC is the alternative, a more standardized concept, with better structured format.
Before removing MI and letting everyone move to RPC, it might be wise to go over all RPC interfaces and fix them to be neat , sane and somewhat consistent. For example, there are still interfaces coding arrays as hashes with multiple identically named fields (htable.dump is one i recently ran into) . There is no language and/or JSON/XML library i know of that can properly handle those (most of them will just end up with the last entry).
In such case the value can be constructed as an array. And I think leaving it at the free will means it won't happen -- we aimed at this since 2008 more or less. Hopefully there are only of few cases that need fixes and can be done while testing 5.0.
But for htable there should not be more than one item with the same name, so such situation should not be for htable.dump. Can you give an example that you encountered, how do you set two items with same name?
Cheers, Daniel
Hello,
comparing the list of rpc vs mi commands, also taking in consideration the interest and expressed feedback from the community, there is no reason to keep MI code longer. Also, during the last days it was signaled that mi_xmlrpc code will need an upgrade for next versions of compilers and xmlrpc-c library, so it's better to get rid of it as soon as possible to avoid work for something to be removed soon.
It should not be any complex work to add the very few missing rpc commands from the mi alternative, so whoever is using a mi command that is not yet available as rpc, should open an issue on bug tracker.
Converting the code from mi to rpc is not complex and everyone is encourage to try converting what they use from mi and it is not a rpc. I can help troubleshooting, provide guidelines, but I have no time to setup test environments for modules that I don use and I don't need those mi/rpc commands.
There is documentation about the RPC API:
- http://www.kamailio.org/docs/docbooks/5.0.x/rpc_api/rpc_api.html
But looking at other modules that have both mi and rpc commands can be a way to learn quickly how to do it.
In several days I plan to start removing MI code, first in the modules that have the rpc commands.
Cheers, Daniel
On 01/12/2016 15:17, Daniel-Constantin Mierla wrote:
Hello,
we started discussing about removing MI (so called management interface) for very long time, more or less since 2008. The RPC should remain the control interface, given its better structure for commands, parameters, etc ... MI is custom protocol using a line-oriented communication via fifo or socket file with kamailio (e.g., implemented mi_fifo and mi_datagram modules). RPC is the alternative, a more standardized concept, with better structured format.
I think it's time to set a clear roadmap for doing the removal. Overall, it will be easier to maintain the code, right now being duplicated code for doing the same operation over MI or RPC, and MI shows its limitations (or complexity to deal with) for advanced needs (see the discussions about how to provide multi-line value parameters over MI).
So, I want to know if there are many relying on MI directly and they still want to keep it, what would be the expected duration they need for upgrading their tools to work with RPC interface, other relevant aspects people have in favour of mi vs rpc.
I am even willing to do the removal in time befire freezing the 5.0 branch. We will ensure a clean start of 5.x series.
The main concern from my point of view is kamctl -- but I think we can preserve the compatibility for kamctl commands and parameters (so command line execution of kamctl will be the same), but the output might be different. That's because it should be easy to updated it to communicate with jsonrpc-s module, but then it will get json-formatted results.
To summarize, two big questions to answer:
a) Are you ok to remove the MI code/commands?
b) If yes to a), are you ok to be done for v5.0?
Not providing feedback will be considered as 'yes' for both questions, so **speak up if you want MI to be kept or delay it removal**.
Cheers, Daniel
-
Alex Balashov -
Alex Hermann -
Brandon Armstead -
Daniel-Constantin Mierla -
jh＠tutpro.com -
Juha Heinanen