Unsubscribing from events in the event callback

# 8 years ago
zreszela	Dear Tango Experts! I have a question reagarding events. In my setup I have a device server (DS1) which exports a device (Dev1) with two attributes (Attr1 and Attr2). In the client I subscribe to the events of Attr1 and Attr2. I would like to unsubscribe from Attr2 events when I receive a specific event from Attr1. I have tried to do it directly in the Attr1 event callback (call unsubscribe_event with id of Attr2 subscription) but it ends with a deadlock. Is it a normal behavior? Thanks for your explanation!

# 8 years ago

Dear Tango Experts!

I have a question reagarding events.

In my setup I have a device server (DS1) which exports a device (Dev1) with two attributes (Attr1 and Attr2). In the client I subscribe to the events of Attr1 and Attr2.
I would like to unsubscribe from Attr2 events when I receive a specific event from Attr1.
I have tried to do it directly in the Attr1 event callback (call unsubscribe_event with id of Attr2 subscription) but it ends with a deadlock. Is it a normal behavior?

Thanks for your explanation!

# 8 years ago
Manu	Hi Zibi, I have made a small C++ client which is doing what you describe. I don't have problem. I attach this small client to this post. Is it what you try to do? Cheers Emmmanuel Attachments: 2Kb

# 8 years ago
zreszela	Hi Manu, Thanks for your your answer and the example. The example is exactly what I want to do, with a small difference that the event type is CHANGE. But I think it does not matter. I have already done some tests with your client and TangoTest. I do not have clear conclussions, my results are very strange: Test 1 OS: openSUSE 13.1, 64 bits Tango: 8.1.2 (with 4 patches from the Tango website) TANGO_HOST and TangoTest and the client run on the same PC Result: deadlock Test 2 OS: openSUSE 11.1, 32 bits Tango: 8.1.2 (with 4 patches from the Tango website + 1 internal patch of ALBA) TANGO_HOST and TangoTest and the client run on the same PC Result: deadlock Test 3 OS: openSUSE 13.1, 64 bits Tango: 8.1.2 (with 4 patches from the Tango website) TANGO_HOST and TangoTest run on another PC (openSUSE 11.1, 32 bits) with Tango 7.2.6 Result: works! Test 4 OS: openSUSE 42.1, 64 bits Tango: 8.1.2 (with 4 patches from the Tango website + 1 internal patch of ALBA) TANGO_HOST and TangoTest run on another PC (openSUSE 11.1, 32 bits) with Tango 7.2.6 Result: works! Today I leave on 3 weeks holidays, so I won't be able to reply regularly on this thread, sorry.. Thanks again, Zibi

# 8 years ago

zreszela

Hi Manu,

Thanks for your your answer and the example. The example is exactly what I want to do, with a small difference that the event type is CHANGE. But I think it does not matter.

I have already done some tests with your client and TangoTest. I do not have clear conclussions, my results are very strange:

Test 1

OS: openSUSE 13.1, 64 bits
Tango: 8.1.2 (with 4 patches from the Tango website)
TANGO_HOST and TangoTest and the client run on the same PC
Result: deadlock

Test 2

OS: openSUSE 11.1, 32 bits
Tango: 8.1.2 (with 4 patches from the Tango website + 1 internal patch of ALBA)
TANGO_HOST and TangoTest and the client run on the same PC
Result: deadlock

Test 3

OS: openSUSE 13.1, 64 bits
Tango: 8.1.2 (with 4 patches from the Tango website)
TANGO_HOST and TangoTest run on another PC (openSUSE 11.1, 32 bits) with Tango 7.2.6
Result: works!

Test 4

OS: openSUSE 42.1, 64 bits
Tango: 8.1.2 (with 4 patches from the Tango website + 1 internal patch of ALBA)
TANGO_HOST and TangoTest run on another PC (openSUSE 11.1, 32 bits) with Tango 7.2.6
Result: works!

Today I leave on 3 weeks holidays, so I won't be able to reply regularly on this thread, sorry..

Thanks again,
Zibi

# 8 years ago
zreszela	Hello again, I have continued with the tests. Below some more results: Test 5 OS: openSUSE 12.1 64 bits Tango: 9.1.0 (with 2 patches from the Tango website) TANGO_HOST and TangoTest run on another PC (openSUSE 11.1 64 bits) with Tango 8.1.2 Result: works! Test 6 OS: openSUSE 13.1 64 bits Tango: 9.1.0 (with 2 patches from the Tango website) TANGO_HOST and TangoTest and the client run on the same PC Result: works! After the first 4 tests I was thinking that maybe it could be related to the configuration where all: DataBaseds, TangoTest and the client run on the same PC. But in the Test 6, using full setup of Tango 9.1.0, there are no problem. Do you think that it could be related to this specific configuration but using the Tango 8.1.2? Do you want me to do some other tests or provide more information? Thanks a lot! Zibi

# 8 years ago

zreszela

Hello again,

I have continued with the tests. Below some more results:

Test 5

OS: openSUSE 12.1 64 bits
Tango: 9.1.0 (with 2 patches from the Tango website)
TANGO_HOST and TangoTest run on another PC (openSUSE 11.1 64 bits) with Tango 8.1.2
Result: works!

Test 6

OS: openSUSE 13.1 64 bits
Tango: 9.1.0 (with 2 patches from the Tango website)
TANGO_HOST and TangoTest and the client run on the same PC
Result: works!

After the first 4 tests I was thinking that maybe it could be related to the configuration where all: DataBaseds, TangoTest and the client run on the same PC. But in the Test 6, using full setup of Tango 9.1.0, there are no problem. Do you think that it could be related to this specific configuration but using the Tango 8.1.2?

Do you want me to do some other tests or provide more information?

Thanks a lot!
Zibi

# 8 years ago
zreszela	Hi, I attach backtraces of all the 6 threads of the client process (when the event callback deadlocks on the unsubscribe_event call) from the Test 1. Cheers, Zibi Attachments: 5Kb

# 8 years ago
Manu	Hi Zibi, The stack traces are interesting. I will look into this and keep you informed. But I thought you are in holidays. In my case, it is this evening until the 5 of January. Wonderful! But no snow in Grenoble! Cheers Manu

# 8 years ago
Manu	Hi Zibi, First the classical sentence these days: Happy new year to you and to everybody at ALBA. From your tests, it seems that Tango 9.1 solves your problem because it works fine for your tests 5 and 6. I checked into the code repository and effectively there is a change in this part of code between Tango 8 and 9. I completely forgot this change and it is not documented in the TANGO_CHANGES file! The code I am referring to is about un-subscription in event callback and more precisely about the detection that the unsubscribe call is done in the callback. When this works fine, the real un-subscription is done by a thread spawn by Tango to avoid the deadlock. In the stack traces you provided, this thread is not there meaning that the detection of event un-subscription during the callback failed. I don't think this could happen with the code used in Tango 9. Do you confirm that using Tango 9 you are not able to re-produce the problem? Cheers Emmanuel

# 8 years ago

Manu

Hi Zibi,

First the classical sentence these days: Happy new year to you and to everybody at ALBA.

From your tests, it seems that Tango 9.1 solves your problem because it works fine for your tests 5 and 6.
I checked into the code repository and effectively there is a change in this part of code between Tango 8 and 9.
I completely forgot this change and it is not documented in the TANGO_CHANGES file!
The code I am referring to is about un-subscription in event callback and more precisely about the detection that the unsubscribe call is done in the callback. When this works fine, the real un-subscription is done by a thread spawn by Tango to avoid the
deadlock. In the stack traces you provided, this thread is not there meaning that the detection of event un-subscription during the callback failed.
I don't think this could happen with the code used in Tango 9.
Do you confirm that using Tango 9 you are not able to re-produce the problem?

Cheers

Emmanuel

# 8 years ago
zreszela	Hello Manu, Happy New Year to you too! I had also spent "strange" Christmans in Poland, with +10 degrees! Thanks for looking into this problem. I confirm that in Tango 9 everything works well. I think your suspicions are right. I have copied the solution from Tango 9 into the Tango 8, and the deadlock disappeared. I think it could be interesting to make a public patch for that (we want to base on that some solutions in Sardana..), what is your opinion? I attach here the changes that solved the problem for me (I was not very creative, just copied the necessary code from the Tango 9). I'm not sure, but I think that the notifd event consumer should also initialize the thread_id. If you agree, I could add this change as well and prepare a new patch. Cheers, Zibi Attachments: 2Kb

# 8 years ago

zreszela

Hello Manu,

Happy New Year to you too! I had also spent "strange" Christmans in Poland, with +10 degrees!

Thanks for looking into this problem. I confirm that in Tango 9 everything works well. I think your suspicions are right.
I have copied the solution from Tango 9 into the Tango 8, and the deadlock disappeared. I think it could be interesting to make a public patch for that (we want to base on that some solutions in Sardana..), what is your opinion?

I attach here the changes that solved the problem for me (I was not very creative, just copied the necessary code from the Tango 9). I'm not sure, but I think that the notifd event consumer should also initialize the thread_id. If you agree, I could add this change as well and prepare a new patch.

Cheers,
Zibi