The elasticsearch changes will I think be the final breaking changes for g6 right?
I’m helping @eikichi18 to develop an application using guillotina_amqp, but we found a problem trying to connect the “task” created by the worker and a simple web socket. We’ve created an object using the guillotina_amqp worker, notified the event to a subscriber, which put it on an utility queue, and tried to sent on a web socket, as shown during the training. The problem is that the worker do not allowed us to open the socket directly so we are force to open a second service on an other port that manages it. So we have the worker service opened with the “g amqp-worker” command, on the local port 5672, and the socket service open with the simple “g -c config.yaml” o “g” command, on the 8080. At this point we find the problem: the worker does not see the list of opened sockets and the socket does not read the events notified by the worker when an object is created in the database, as if we are managing two different instance. We think it could be a solution to save the socket’s list in the context using a new behavior, but first we are glad to hear a more expert opinion, maybe there is an easier way to connect the socket and the worker and we are doing something wrong.
regarding @Samuele575 's question, not sure I understand completely.
But if you have the separate service handling websocket connections, the worker task should be able to send some notifications too, no?
Probably you can't pass the open websocket connection to the worker as a parameter, because the parameters are json-serialized before sending to rabbitmq, but maybe you could pass the worker ws url?
hope it helped
I’ll try to explain better the problem from another prospective: we are unable to report events generated by the worker to the socket manager. The @task use two functions: create_content_in_container and notify, so the subscriber can handle events but only the worker service reads them and starts the subscribers, the socket manager does not.
I was working on the guillotina object server (POC) which stores transactional objects inside rocksdb and it is using grpc.
Have a look https://github.com/kakshay21/gobject - ignore service name ;)
I also need little help in integrating with guiilotina, as currently, I'm just storing hardcoded values from python grpc client.
- plone/guillotina#954 ( kind of bad how this can affect the way a browser will translate a path)
- plone/guillotina#955
- plone/guillotina#956
I was reading getting started guide on guillotina and while I was making any change in the endpoint, I had to stop the server, install the package
pip install -e mypackage and then starting the server. Doesn't guiilotina reload on its own on every code change like django or rails?
1 reply
awesome-asgi repo to add Guillotina florimondmanca/awesome-asgi#41. It needs 20 upvotes to be merged... Can you help with a :+1: ? :)
@pigeonflight I think I found the problem: plone/guillotina@34c6bc0 In G6 we changed some methods of the ICatalog interface and the behavior of the catalog api.
GET @search uses ICatalog.search and expects queries to be parsed
POST @search uses ICatalog.search_raw and expects raw queries (i.e. ES queries)
The former is not supported in pgcatalog. So, I'd say it's not a bug and could be solved changing the api call to a GET. Maybe this breaking change could have had been documented better
With a simple reservation calendar thus:
class ICalendar(Interface):
name = TextLine()
@contenttype(type_name="Calendar", schema=ICalendar, behaviors=[DCbehavior])
class ReservationCalendar(content.Folder):
"calendar resource type”.. and then POSTing a new calendar with JSON body:
{
"@type": "Calendar",
"name": "Test calendar"
}the response is:
{
"@id": "http://localhost:8080/db/calendars/d9a348011fea46d5baab516fdd534941",
"@name": "d9a348011fea46d5baab516fdd534941",
"@type": "Calendar",
"@uid": "c19|d9a348011fea46d5baab516fdd534941"
}Why is the calendar name not returned? As far as I can tell, this is 100% according to how docs instruct? Is name a reserved word because @name, or something?
name as well
@pigeonflight I think I found the problem: plone/guillotina@34c6bc0 In G6 we changed some methods of the ICatalog interface and the behavior of the catalog api.
GET @search uses
ICatalog.searchand expects queries to be parsed
POST @search usesICatalog.search_rawand expects raw queries (i.e. ES queries)The former is not supported in pgcatalog. So, I'd say it's not a bug and could be solved changing the api call to a GET. Maybe this breaking change could have had been documented better
I need to re-read this to understand the implications. Are you just saying that GET based search is off the table when using pgcatalog?