[ovs-dev] [PATCH] ovn-controller: Fix flow installation latency caused by recompute.

Mark Michelson mmichels at redhat.com
Thu Jul 25 20:23:01 UTC 2019


Forgot the e-mail address after my name :)

Acked-by: Mark Michelson <mmichels at redhat.com>

On 7/25/19 4:16 PM, Mark Michelson wrote:
> Acked-by: Mark Michelson
> 
> Thanks for the informative commit message!
> 
> On 7/23/19 10:52 PM, Han Zhou wrote:
>> From: Han Zhou <hzhou8 at ebay.com>
>>
>> When there are in-flight flow-installations pending to ovs-vswitchd,
>> current incremental processing logic prioritizes new change handling.
>> However, in scenarios where frequent recomputes are triggered, the
>> later recompute would block the flow-installation for previously
>> computed flows because recompute usually takes long time, especially
>> when there are large number of flows. This results in worse latency
>> than the version without incremental processing in specific scale
>> test scenarios.
>>
>> While we can simply fix the problem by prioritizing flow installation
>> rather than new change handling, it can cause the incremental
>> processing to degrade to always recompute in certain scenarios when
>> there are some changes triggering recomputes, followed by a lot of
>> continously coming changes that can be handled incrementally. Because
>> OVSDB change tracker cannot preserve changes across iterations, once
>> the recompute is triggered and resulted in a lot of pending messages
>> to ovs-vswitchd, and if we choose to skip the engine_run()
>> in the next iteration when a incrementally processible change comes,
>> we miss the opportunity to handle that tracked change and will have
>> to trigger recompute again in the next next iteration, and so on, if
>> such changes come continously.
>>
>> This patch solves the problem by introducing 
>> engine_set_abort_recompute(),
>> so that we can prioritize new change handling if the change can be
>> incrementally processed, but if the change triggers recompute, we
>> abort there without spending CPU on the recompute to avoid blocking
>> the previous computed flow installation.
>>
>> Reported-by: Daniel Alvarez Sanchez <dalvarez at redhat.com>
>> Reported-by: Numan Siddique <nusiddiq at redhat.com>
>> Reported-at: 
>> https://mail.openvswitch.org/pipermail/ovs-discuss/2019-June/048822.html
>> Tested-by: Numan Siddique <nusiddiq at redhat.com>
>> Signed-off-by: Han Zhou <hzhou8 at ebay.com>
>> ---
>>   ovn/controller/ofctrl.c         |  2 +-
>>   ovn/controller/ofctrl.h         |  1 +
>>   ovn/controller/ovn-controller.c | 30 +++++++++++++++++++++++++++---
>>   ovn/lib/inc-proc-eng.c          | 26 ++++++++++++++++++++++----
>>   ovn/lib/inc-proc-eng.h          |  9 +++++++--
>>   5 files changed, 58 insertions(+), 10 deletions(-)
>>
>> diff --git a/ovn/controller/ofctrl.c b/ovn/controller/ofctrl.c
>> index 043abd6..0fcaa72 100644
>> --- a/ovn/controller/ofctrl.c
>> +++ b/ovn/controller/ofctrl.c
>> @@ -985,7 +985,7 @@ add_meter(struct ovn_extend_table_info *m_desired,
>>    * in the correct state and not backlogged with existing flow_mods.  
>> (Our
>>    * criteria for being backlogged appear very conservative, but the 
>> socket
>>    * between ovn-controller and OVS provides some buffering.) */
>> -static bool
>> +bool
>>   ofctrl_can_put(void)
>>   {
>>       if (state != S_UPDATE_FLOWS
>> diff --git a/ovn/controller/ofctrl.h b/ovn/controller/ofctrl.h
>> index ed8918a..2b21c11 100644
>> --- a/ovn/controller/ofctrl.h
>> +++ b/ovn/controller/ofctrl.h
>> @@ -51,6 +51,7 @@ void ofctrl_put(struct ovn_desired_flow_table *,
>>                   const struct sbrec_meter_table *,
>>                   int64_t nb_cfg,
>>                   bool flow_changed);
>> +bool ofctrl_can_put(void);
>>   void ofctrl_wait(void);
>>   void ofctrl_destroy(void);
>>   int64_t ofctrl_get_cur_cfg(void);
>> diff --git a/ovn/controller/ovn-controller.c 
>> b/ovn/controller/ovn-controller.c
>> index c4883aa..866cc1c 100644
>> --- a/ovn/controller/ovn-controller.c
>> +++ b/ovn/controller/ovn-controller.c
>> @@ -1869,6 +1869,7 @@ main(int argc, char *argv[])
>>       uint64_t engine_run_id = 0;
>>       uint64_t old_engine_run_id = 0;
>> +    bool engine_aborted = false;
>>       unsigned int ovs_cond_seqno = UINT_MAX;
>>       unsigned int ovnsb_cond_seqno = UINT_MAX;
>> @@ -1955,7 +1956,30 @@ main(int argc, char *argv[])
>>                       stopwatch_start(CONTROLLER_LOOP_STOPWATCH_NAME,
>>                                       time_msec());
>>                       if (ovnsb_idl_txn) {
>> -                        engine_run(&en_flow_output, ++engine_run_id);
>> +                        if (!ofctrl_can_put()) {
>> +                            /* When there are in-flight messages 
>> pending to
>> +                             * ovs-vswitchd, we should hold on 
>> recomputing so
>> +                             * that the previous flow installations 
>> won't be
>> +                             * delayed.  However, we still want to 
>> try if
>> +                             * recompute is not needed and we can 
>> quickly
>> +                             * incrementally process the new changes, 
>> to avoid
>> +                             * unnecessarily forced recomputes later 
>> on.  This
>> +                             * is because the OVSDB change tracker 
>> cannot
>> +                             * preserve tracked changes across 
>> iterations.  If
>> +                             * change tracking is improved, we can 
>> simply skip
>> +                             * this round of engine_run and continue 
>> processing
>> +                             * acculated changes incrementally later 
>> when
>> +                             * ofctrl_can_put() returns true. */
>> +                            if (!engine_aborted) {
>> +                                engine_set_abort_recompute(true);
>> +                                engine_aborted = 
>> engine_run(&en_flow_output,
>> +                                                            
>> ++engine_run_id);
>> +                            }
>> +                        } else {
>> +                            engine_set_abort_recompute(false);
>> +                            engine_aborted = false;
>> +                            engine_run(&en_flow_output, 
>> ++engine_run_id);
>> +                        }
>>                       }
>>                       stopwatch_stop(CONTROLLER_LOOP_STOPWATCH_NAME,
>>                                      time_msec());
>> @@ -1993,8 +2017,8 @@ main(int argc, char *argv[])
>>                   }
>>               }
>> -            if (old_engine_run_id == engine_run_id) {
>> -                if (engine_need_run(&en_flow_output)) {
>> +            if (old_engine_run_id == engine_run_id || engine_aborted) {
>> +                if (engine_aborted || 
>> engine_need_run(&en_flow_output)) {
>>                       VLOG_DBG("engine did not run, force recompute 
>> next time: "
>>                                "br_int %p, chassis %p", br_int, chassis);
>>                       engine_set_force_recompute(true);
>> diff --git a/ovn/lib/inc-proc-eng.c b/ovn/lib/inc-proc-eng.c
>> index 1ddea1a..1064a08 100644
>> --- a/ovn/lib/inc-proc-eng.c
>> +++ b/ovn/lib/inc-proc-eng.c
>> @@ -31,6 +31,7 @@
>>   VLOG_DEFINE_THIS_MODULE(inc_proc_eng);
>>   static bool engine_force_recompute = false;
>> +static bool engine_abort_recompute = false;
>>   static const struct engine_context *engine_context;
>>   void
>> @@ -39,6 +40,12 @@ engine_set_force_recompute(bool val)
>>       engine_force_recompute = val;
>>   }
>> +void
>> +engine_set_abort_recompute(bool val)
>> +{
>> +    engine_abort_recompute = val;
>> +}
>> +
>>   const struct engine_context *
>>   engine_get_context(void)
>>   {
>> @@ -121,11 +128,11 @@ engine_ovsdb_node_add_index(struct engine_node 
>> *node, const char *name,
>>       ed->n_indexes ++;
>>   }
>> -void
>> +bool
>>   engine_run(struct engine_node *node, uint64_t run_id)
>>   {
>>       if (node->run_id == run_id) {
>> -        return;
>> +        return true;
>>       }
>>       node->run_id = run_id;
>> @@ -133,11 +140,13 @@ engine_run(struct engine_node *node, uint64_t 
>> run_id)
>>       if (!node->n_inputs) {
>>           node->run(node);
>>           VLOG_DBG("node: %s, changed: %d", node->name, node->changed);
>> -        return;
>> +        return true;
>>       }
>>       for (size_t i = 0; i < node->n_inputs; i++) {
>> -        engine_run(node->inputs[i].node, run_id);
>> +        if (!engine_run(node->inputs[i].node, run_id)) {
>> +            return false;
>> +        }
>>       }
>>       bool need_compute = false;
>> @@ -160,6 +169,10 @@ engine_run(struct engine_node *node, uint64_t 
>> run_id)
>>       if (need_recompute) {
>>           VLOG_DBG("node: %s, recompute (%s)", node->name,
>>                    engine_force_recompute ? "forced" : "triggered");
>> +        if (engine_abort_recompute) {
>> +            VLOG_DBG("node: %s, recompute aborted", node->name);
>> +            return false;
>> +        }
>>           node->run(node);
>>       } else if (need_compute) {
>>           for (size_t i = 0; i < node->n_inputs; i++) {
>> @@ -170,6 +183,10 @@ engine_run(struct engine_node *node, uint64_t 
>> run_id)
>>                       VLOG_DBG("node: %s, can't handle change for 
>> input %s, "
>>                                "fall back to recompute",
>>                                node->name, node->inputs[i].node->name);
>> +                    if (engine_abort_recompute) {
>> +                        VLOG_DBG("node: %s, recompute aborted", 
>> node->name);
>> +                        return false;
>> +                    }
>>                       node->run(node);
>>                       break;
>>                   }
>> @@ -178,6 +195,7 @@ engine_run(struct engine_node *node, uint64_t run_id)
>>       }
>>       VLOG_DBG("node: %s, changed: %d", node->name, node->changed);
>> +    return true;
>>   }
>>   bool
>> diff --git a/ovn/lib/inc-proc-eng.h b/ovn/lib/inc-proc-eng.h
>> index aab899e..c3d7b5e 100644
>> --- a/ovn/lib/inc-proc-eng.h
>> +++ b/ovn/lib/inc-proc-eng.h
>> @@ -121,8 +121,9 @@ struct engine_node {
>>   void engine_init(struct engine_node *);
>>   /* Execute the processing recursively, which should be called in the 
>> main
>> - * loop. */
>> -void engine_run(struct engine_node *, uint64_t run_id);
>> + * loop. Returns true if the execution is compelte, false if it is 
>> aborted,
>> + * which could happen when engine_abort_recompute is set. */
>> +bool engine_run(struct engine_node *, uint64_t run_id);
>>   /* Clean up the data for the engine nodes recursively. It calls each 
>> node's
>>    * cleanup() method if not NULL. It should be called before the program
>> @@ -150,6 +151,10 @@ void engine_add_input(struct engine_node *node, 
>> struct engine_node *input,
>>    * iteration, and the change can't be tracked across iterations */
>>   void engine_set_force_recompute(bool val);
>> +/* Set the flag to cause engine execution to be aborted when there
>> + * is any recompute to be triggered in any node. */
>> +void engine_set_abort_recompute(bool val);
>> +
>>   const struct engine_context * engine_get_context(void);
>>   void engine_set_context(const struct engine_context *);
>>
> 



More information about the dev mailing list