[Tarantool-patches] [PATCH 1/1] test: fix flaky box/gh-4627-session-use-after-free

[Tarantool-patches] [PATCH 1/1] test: fix flaky box/gh-4627-session-use-after-free

[Vladislav Shpilevoy]

The problem was in that the test uses the global trigger
box.session.on_disconnect() to set a global variable by one
connection. But test-run can do multiple connects/reconnects to
the same instance. That led to multiple invocations of
box.session.on_disconnect(), which could override the global
variable in unexpected ways and moments.

The patch makes only one session execute that trigger.

Probably related to https://github.com/tarantool/test-run/issues/46
Follow up #4627
---
Branch: https://github.com/tarantool/tarantool/tree/gerold103/session-flaky-test

 .../box/gh-4627-session-use-after-free.result | 28 +++++++++++++++++++
 .../gh-4627-session-use-after-free.test.lua   | 20 +++++++++++++
 2 files changed, 48 insertions(+)

diff --git a/test/box/gh-4627-session-use-after-free.result b/test/box/gh-4627-session-use-after-free.result
index 5e5c154b9..33e5a7af1 100644
--- a/test/box/gh-4627-session-use-after-free.result
+++ b/test/box/gh-4627-session-use-after-free.result
@@ -20,6 +20,21 @@ fiber = require('fiber')
  | ---
  | ...
 
+box.schema.user.grant('guest', 'execute', 'universe')
+ | ---
+ | ...
+
+-- This is a workaround for flakiness of the test
+-- appearing when another test tries to connect to
+-- this one. By setting a flag for only one
+-- session, others won't interfere in
+-- session.on_disconnect().
+function enable_on_disconnect()                 \
+    box.session.storage.is_enabled = true       \
+end
+ | ---
+ | ...
+
 sid_before_yield = nil
  | ---
  | ...
@@ -27,6 +42,9 @@ sid_after_yield = nil
  | ---
  | ...
 func = box.session.on_disconnect(function()     \
+    if not box.session.storage.is_enabled then  \
+        return                                  \
+    end                                         \
     sid_before_yield = box.session.id()         \
     fiber.yield()                               \
     sid_after_yield = box.session.id()          \
@@ -37,6 +55,12 @@ end)
 connection = net_box.connect(box.cfg.listen)
  | ---
  | ...
+-- Connections, not related to this one, won't
+-- call this function, and therefore won't do
+-- anything in session.on_disconnect() trigger.
+connection:call('enable_on_disconnect')
+ | ---
+ | ...
 connection:ping()
  | ---
  | - true
@@ -58,3 +82,7 @@ sid_after_yield == sid_before_yield and sid_after_yield ~= 0 or \
 box.session.on_disconnect(nil, func)
  | ---
  | ...
+
+box.schema.user.revoke('guest', 'execute', 'universe')
+ | ---
+ | ...
diff --git a/test/box/gh-4627-session-use-after-free.test.lua b/test/box/gh-4627-session-use-after-free.test.lua
index 70624a96a..1b2ebfdc4 100644
--- a/test/box/gh-4627-session-use-after-free.test.lua
+++ b/test/box/gh-4627-session-use-after-free.test.lua
@@ -15,15 +15,33 @@
 net_box = require('net.box')
 fiber = require('fiber')
 
+box.schema.user.grant('guest', 'execute', 'universe')
+
+-- This is a workaround for flakiness of the test
+-- appearing when another test tries to connect to
+-- this one. By setting a flag for only one
+-- session, others won't interfere in
+-- session.on_disconnect().
+function enable_on_disconnect()                 \
+    box.session.storage.is_enabled = true       \
+end
+
 sid_before_yield = nil
 sid_after_yield = nil
 func = box.session.on_disconnect(function()     \
+    if not box.session.storage.is_enabled then  \
+        return                                  \
+    end                                         \
     sid_before_yield = box.session.id()         \
     fiber.yield()                               \
     sid_after_yield = box.session.id()          \
 end)
 
 connection = net_box.connect(box.cfg.listen)
+-- Connections, not related to this one, won't
+-- call this function, and therefore won't do
+-- anything in session.on_disconnect() trigger.
+connection:call('enable_on_disconnect')
 connection:ping()
 connection:close()
 
@@ -33,3 +51,5 @@ sid_after_yield == sid_before_yield and sid_after_yield ~= 0 or \
     {sid_after_yield, sid_before_yield}
 
 box.session.on_disconnect(nil, func)
+
+box.schema.user.revoke('guest', 'execute', 'universe')
-- 
2.21.0 (Apple Git-122.2)

Re: [Tarantool-patches] [PATCH 1/1] test: fix flaky box/gh-4627-session-use-after-free

[Alexander Turenko]

This fix itself is okay for me, but, please, look at a few minor
comments below.

WBR, Alexander Turenko.

On Tue, Dec 03, 2019 at 01:05:30AM +0100, Vladislav Shpilevoy wrote:
> The problem was in that the test uses the global trigger
> box.session.on_disconnect() to set a global variable by one
> connection. But test-run can do multiple connects/reconnects to
> the same instance. That led to multiple invocations of
> box.session.on_disconnect(), which could override the global
> variable in unexpected ways and moments.
> 
> The patch makes only one session execute that trigger.
> 
> Probably related to https://github.com/tarantool/test-run/issues/46
> Follow up #4627
> ---
> Branch: https://github.com/tarantool/tarantool/tree/gerold103/session-flaky-test
> 
>  .../box/gh-4627-session-use-after-free.result | 28 +++++++++++++++++++
>  .../gh-4627-session-use-after-free.test.lua   | 20 +++++++++++++
>  2 files changed, 48 insertions(+)

> diff --git a/test/box/gh-4627-session-use-after-free.test.lua b/test/box/gh-4627-session-use-after-free.test.lua
> index 70624a96a..1b2ebfdc4 100644
> --- a/test/box/gh-4627-session-use-after-free.test.lua
> +++ b/test/box/gh-4627-session-use-after-free.test.lua
> @@ -15,15 +15,33 @@
>  net_box = require('net.box')
>  fiber = require('fiber')
>  
> +box.schema.user.grant('guest', 'execute', 'universe')
> +
> +-- This is a workaround for flakiness of the test
> +-- appearing when another test tries to connect to

I tentative that another test may access to a unix socket (tarantool's
binary port) that run on another worker: all those sockets are placed in
worker's directory. It seems that test-run itself doing some reconnects.

> +-- this one. By setting a flag for only one
> +-- session, others won't interfere in
> +-- session.on_disconnect().
> +function enable_on_disconnect()                 \
> +    box.session.storage.is_enabled = true       \
> +end
> +
>  sid_before_yield = nil
>  sid_after_yield = nil
>  func = box.session.on_disconnect(function()     \
> +    if not box.session.storage.is_enabled then  \
> +        return                                  \
> +    end                                         \
>      sid_before_yield = box.session.id()         \
>      fiber.yield()                               \
>      sid_after_yield = box.session.id()          \
>  end)
>  
>  connection = net_box.connect(box.cfg.listen)
> +-- Connections, not related to this one, won't
> +-- call this function, and therefore won't do
> +-- anything in session.on_disconnect() trigger.
> +connection:call('enable_on_disconnect')
>  connection:ping()

It seems that :ping() is not needed anymore.

>  connection:close()
>  
> @@ -33,3 +51,5 @@ sid_after_yield == sid_before_yield and sid_after_yield ~= 0 or \
>      {sid_after_yield, sid_before_yield}
>  
>  box.session.on_disconnect(nil, func)
> +
> +box.schema.user.revoke('guest', 'execute', 'universe')
> -- 
> 2.21.0 (Apple Git-122.2)
> 

Re: [Tarantool-patches] [PATCH 1/1] test: fix flaky box/gh-4627-session-use-after-free

[Vladislav Shpilevoy]

Hi! Thanks for the review!

I submitted fixes to the branch.

On 04/12/2019 12:35, Alexander Turenko wrote:
> This fix itself is okay for me, but, please, look at a few minor
> comments below.
> 
> WBR, Alexander Turenko.
> 
> On Tue, Dec 03, 2019 at 01:05:30AM +0100, Vladislav Shpilevoy wrote:
>> The problem was in that the test uses the global trigger
>> box.session.on_disconnect() to set a global variable by one
>> connection. But test-run can do multiple connects/reconnects to
>> the same instance. That led to multiple invocations of
>> box.session.on_disconnect(), which could override the global
>> variable in unexpected ways and moments.
>>
>> The patch makes only one session execute that trigger.
>>
>> Probably related to https://github.com/tarantool/test-run/issues/46
>> Follow up #4627
>> ---
>> Branch: https://github.com/tarantool/tarantool/tree/gerold103/session-flaky-test
>>
>>  .../box/gh-4627-session-use-after-free.result | 28 +++++++++++++++++++
>>  .../gh-4627-session-use-after-free.test.lua   | 20 +++++++++++++
>>  2 files changed, 48 insertions(+)
> 
>> diff --git a/test/box/gh-4627-session-use-after-free.test.lua b/test/box/gh-4627-session-use-after-free.test.lua
>> index 70624a96a..1b2ebfdc4 100644
>> --- a/test/box/gh-4627-session-use-after-free.test.lua
>> +++ b/test/box/gh-4627-session-use-after-free.test.lua
>> @@ -15,15 +15,33 @@
>>  net_box = require('net.box')
>>  fiber = require('fiber')
>>  
>> +box.schema.user.grant('guest', 'execute', 'universe')
>> +
>> +-- This is a workaround for flakiness of the test
>> +-- appearing when another test tries to connect to
> 
> I tentative that another test may access to a unix socket (tarantool's

I am sorry, but I don't understand - what do you mean as 'I tentative'?
This is not a verb. You meant, that you guess, assume, doubt, investigated
or something else?

I assume you've meant 'I doubt', as so there is a fix:

===========================================================================

diff --git a/test/box/gh-4627-session-use-after-free.result b/test/box/gh-4627-session-use-after-free.result
index 33e5a7af1..67196b33d 100644
--- a/test/box/gh-4627-session-use-after-free.result
+++ b/test/box/gh-4627-session-use-after-free.result
@@ -25,8 +25,9 @@ box.schema.user.grant('guest', 'execute', 'universe')
  | ...
 
 -- This is a workaround for flakiness of the test
--- appearing when another test tries to connect to
--- this one. By setting a flag for only one
+-- appearing when test-run does reconnects to the
+-- instance and therefore creates multiple
+-- sessions. By setting a flag for only one
 -- session, others won't interfere in
 -- session.on_disconnect().
 function enable_on_disconnect()                 \
diff --git a/test/box/gh-4627-session-use-after-free.test.lua b/test/box/gh-4627-session-use-after-free.test.lua
index 1b2ebfdc4..256195186 100644
--- a/test/box/gh-4627-session-use-after-free.test.lua
+++ b/test/box/gh-4627-session-use-after-free.test.lua
@@ -18,8 +18,9 @@ fiber = require('fiber')
 box.schema.user.grant('guest', 'execute', 'universe')
 
 -- This is a workaround for flakiness of the test
--- appearing when another test tries to connect to
--- this one. By setting a flag for only one
+-- appearing when test-run does reconnects to the
+-- instance and therefore creates multiple
+-- sessions. By setting a flag for only one
 -- session, others won't interfere in
 -- session.on_disconnect().
 function enable_on_disconnect()                 \

===========================================================================

> binary port) that run on another worker: all those sockets are placed in
> worker's directory. It seems that test-run itself doing some reconnects.
> 
>> +-- this one. By setting a flag for only one
>> +-- session, others won't interfere in
>> +-- session.on_disconnect().
>> +function enable_on_disconnect()                 \
>> +    box.session.storage.is_enabled = true       \
>> +end
>> +
>>  sid_before_yield = nil
>>  sid_after_yield = nil
>>  func = box.session.on_disconnect(function()     \
>> +    if not box.session.storage.is_enabled then  \
>> +        return                                  \
>> +    end                                         \
>>      sid_before_yield = box.session.id()         \
>>      fiber.yield()                               \
>>      sid_after_yield = box.session.id()          \
>>  end)
>>  
>>  connection = net_box.connect(box.cfg.listen)
>> +-- Connections, not related to this one, won't
>> +-- call this function, and therefore won't do
>> +-- anything in session.on_disconnect() trigger.
>> +connection:call('enable_on_disconnect')
>>  connection:ping()
> 
> It seems that :ping() is not needed anymore.
> 

Yes, it is not.

===========================================================================

diff --git a/test/box/gh-4627-session-use-after-free.result b/test/box/gh-4627-session-use-after-free.result
index 33e5a7af1..29144b098 100644
--- a/test/box/gh-4627-session-use-after-free.result
+++ b/test/box/gh-4627-session-use-after-free.result
@@ -61,10 +62,6 @@ connection = net_box.connect(box.cfg.listen)
 connection:call('enable_on_disconnect')
  | ---
  | ...
-connection:ping()
- | ---
- | - true
- | ...
 connection:close()
  | ---
  | ...
diff --git a/test/box/gh-4627-session-use-after-free.test.lua b/test/box/gh-4627-session-use-after-free.test.lua
index 1b2ebfdc4..18d56e203 100644
--- a/test/box/gh-4627-session-use-after-free.test.lua
+++ b/test/box/gh-4627-session-use-after-free.test.lua
@@ -42,7 +43,6 @@ connection = net_box.connect(box.cfg.listen)
 -- call this function, and therefore won't do
 -- anything in session.on_disconnect() trigger.
 connection:call('enable_on_disconnect')
-connection:ping()
 connection:close()
 
 while not sid_after_yield do fiber.yield() end

===========================================================================

Re: [Tarantool-patches] [PATCH 1/1] test: fix flaky box/gh-4627-session-use-after-free

[Alexander Turenko]

LGTM.

Pushed to master, 2.2 and 1.10.

CCed Kirill.

WBR, Alexander Turenko.

On Thu, Dec 05, 2019 at 09:42:11PM +0100, Vladislav Shpilevoy wrote:
> Hi! Thanks for the review!
> 
> I submitted fixes to the branch.

> >> +-- This is a workaround for flakiness of the test
> >> +-- appearing when another test tries to connect to
> > 
> > I tentative that another test may access to a unix socket (tarantool's
> 
> I am sorry, but I don't understand - what do you mean as 'I tentative'?
> This is not a verb. You meant, that you guess, assume, doubt, investigated
> or something else?
> 
> I assume you've meant 'I doubt', as so there is a fix:

Exactly.