issue: 2247939 Add option VMA_VMAD_ENABLED #907
Conversation
|
Test PASSed. |
There was a problem hiding this comment.
Please consider checking for g_p_agent == NULL instead of safe_mce_sys().vmad_enabled.
It looks as more universal way that independent from approach if VMA has special env variable or not.
In this case it is enough to handle enable/disable just in one place in main().
Probably additional check is redundant for ring_tap because warning should be appeared once on initialization when g_p_agent is initialized on not (see possible warning at output_fatal())
This option allows to disable vmad functionality in VMA. Running vmad process can lead to performance degradation in case of multiple TCP socket state changes. Signed-off-by: Dmytro Podgornyi <[email protected]>
|
Test PASSed. |
| @@ -371,6 +371,9 @@ int ring_tap::prepare_flow_message(vma_msg_flow& data, msg_flow_t flow_action, | |||
| { | |||
| int rc = 0; | |||
|
|
|||
| if (unlikely(g_p_agent == NULL)) | |||
| return 0; | |||
|
|
|||
There was a problem hiding this comment.
it is needless and incorrect because it conflicts with
/* create egress rule (redirect traffic from tap device to physical interface) */
rc = prepare_flow_message(data, VMA_MSG_FLOW_ADD);
if (rc != 0) {
ring_logwarn("Add TC rule failed with error=%d", rc);
}
ring_tap() does not work correctly w/o vma daemon.
So preferable way is to report error once ring_tap() is called. It can be before call or inside constructor.
| @@ -400,6 +403,9 @@ int ring_tap::prepare_flow_message(vma_msg_flow& data, msg_flow_t flow_action) | |||
| { | |||
| int rc = 0; | |||
|
|
|||
| if (unlikely(g_p_agent == NULL)) | |||
| return 0; | |||
|
Can one of the admins verify this patch? |
This option allows to disable vmad functionality in VMA. Running vmad process
can lead to performance degradation in case of multiple TCP socket state
changes.