[chassis][syncd][sai] Adjusting response timeout during syncd init (s…

…onic-net#2159) In VOQ based chassis where syncd uses VOQ SAI, if there are large number of front panel ports, SAI takes more than 1 minutes to complete the switch create initialization. Because of this, the switch create request sent by orchagent is not getting response within the default response wait time of 1 minute. So the orchagent declares switch create failure and crashes. To fix this, in orchagent, the syncd response time out is set to 5 minutes for line (voq) card and 10 minutes for supervisor (fabric) card before sending request for switch create and is set back to default wait time after the switch create. Signed-off-by: vedganes <[email protected]>
stephenxs · Mar 1, 2022 · 6e5ed1c · 6e5ed1c
1 parent 0a99f54
commit 6e5ed1c
Showing 1 changed file with 46 additions and 0 deletions.
diff --git a/orchagent/main.cpp b/orchagent/main.cpp
@@ -574,6 +574,36 @@ int main(int argc, char **argv)
     attr.value.u64 = gSwitchId;
     attrs.push_back(attr);
 
+    if (gMySwitchType == "voq" || gMySwitchType == "fabric")
+    {
+        /* We set this long timeout in order for orchagent to wait enough time for
+         * response from syncd. It is needed since switch create takes more time
+         * than default time to create switch if there are lots of front panel ports
+         * and systems ports to initialize
+         */
+
+        if (gMySwitchType == "voq")
+        {
+            attr.value.u64 = (5 * SAI_REDIS_DEFAULT_SYNC_OPERATION_RESPONSE_TIMEOUT);
+        }
+        else if (gMySwitchType == "fabric")
+        {
+            attr.value.u64 = (10 * SAI_REDIS_DEFAULT_SYNC_OPERATION_RESPONSE_TIMEOUT);
+        }
+
+        attr.id = SAI_REDIS_SWITCH_ATTR_SYNC_OPERATION_RESPONSE_TIMEOUT;
+        status = sai_switch_api->set_switch_attribute(gSwitchId, &attr);
+
+        if (status != SAI_STATUS_SUCCESS)
+        {
+            SWSS_LOG_WARN("Failed to set SAI REDIS response timeout");
+        }
+        else
+        {
+            SWSS_LOG_NOTICE("SAI REDIS response timeout set successfully to %" PRIu64 " ", attr.value.u64);
+        }
+    }
+
     status = sai_switch_api->create_switch(&gSwitchId, (uint32_t)attrs.size(), attrs.data());
     if (status != SAI_STATUS_SUCCESS)
     {
@@ -582,6 +612,22 @@ int main(int argc, char **argv)
     }
     SWSS_LOG_NOTICE("Create a switch, id:%" PRIu64, gSwitchId);
 
+    if (gMySwitchType == "voq" || gMySwitchType == "fabric")
+    {
+        /* Set syncd response timeout back to the default value */
+        attr.id = SAI_REDIS_SWITCH_ATTR_SYNC_OPERATION_RESPONSE_TIMEOUT;
+        attr.value.u64 = SAI_REDIS_DEFAULT_SYNC_OPERATION_RESPONSE_TIMEOUT;
+        status = sai_switch_api->set_switch_attribute(gSwitchId, &attr);
+
+        if (status != SAI_STATUS_SUCCESS)
+        {
+            SWSS_LOG_WARN("Failed to set SAI REDIS response timeout to default");
+        }
+        else
+        {
+            SWSS_LOG_NOTICE("SAI REDIS response timeout set successfully to default: %" PRIu64 " ", attr.value.u64);
+        }
+    }
 
     if (gMySwitchType != "fabric")
     {