Re: [PATCH] fetch-pack: add tracing for negotiation rounds

[Josh Steadmon <steadmon@google.com>]

On 2022.07.25 16:07, Junio C Hamano wrote:
> Josh Steadmon <steadmon@google.com> writes:
> 
> > Currently, negotiation for V0/V1/V2 fetch have trace2 regions covering
> > the entire negotiation process. However, we'd like additional data, such
> > as timing for each round of negotiation or the number of "haves" in each
> > round. Additionally, "independent negotiation" (AKA push negotiation)
> > has no tracing at all. Having this data would allow us to compare the
> > performance of the various negotation implementations, and to debug
> > unexpectedly slow fetch & push sessions.
> 
> Quite sensibly argued.  I do not necessarily see the current code as
> "broken", and "fix" at the beginning of the next line may not be an
> appropriate word to describe this enhancement, but I think it would
> be nice to have such numbers.

Reworded that sentence.


> > diff --git a/fetch-pack.c b/fetch-pack.c
> > index cb6647d657..01a451e456 100644
> > --- a/fetch-pack.c
> > +++ b/fetch-pack.c
> > @@ -299,6 +299,7 @@ static int find_common(struct fetch_negotiator *negotiator,
> >  {
> >  	int fetching;
> >  	int count = 0, flushes = 0, flush_at = INITIAL_FLUSH, retval;
> > +	int negotiation_round = 0, haves = 0;
> >  	const struct object_id *oid;
> >  	unsigned in_vain = 0;
> >  	int got_continue = 0;
> > @@ -441,9 +442,19 @@ static int find_common(struct fetch_negotiator *negotiator,
> >  		packet_buf_write(&req_buf, "have %s\n", oid_to_hex(oid));
> >  		print_verbose(args, "have %s", oid_to_hex(oid));
> >  		in_vain++;
> > +		haves++;
> >  		if (flush_at <= ++count) {
> >  			int ack;
> >  
> > +			negotiation_round++;
> > +			trace2_region_enter_printf("negotiation_v0_v1", "round",
> > +						   the_repository, "round-%d",
> > +						   negotiation_round);
> 
> Not an objection, but all the hits to existing calls to this
> function show the data in a format as vanilla as possible without
> frills (presumably to make it easier to mechanically parse the value
> out when needed???), and the "round-" prefix we see here somehow
> looks out of place.  Doesn't the fact that the record is in the
> "negotiation_v0_v1" category and has label "round" clear enough sign
> that the value presented is the negotiation_round?

True, fixed in V2.


> > +			trace2_data_intmax("negotiation_v0_v1", the_repository,
> > +					   "haves_added", haves);
> > +			trace2_data_intmax("negotiation_v0_v1", the_repository,
> > +					   "in_vain", in_vain);
> > +			haves = 0;
> >  			packet_buf_flush(&req_buf);
> >  			send_request(args, fd[1], &req_buf);
> >  			strbuf_setlen(&req_buf, state_len);
> > @@ -465,6 +476,9 @@ static int find_common(struct fetch_negotiator *negotiator,
> >  						      ack, oid_to_hex(result_oid));
> >  				switch (ack) {
> >  				case ACK:
> > +					trace2_region_leave_printf("negotiation_v0_v1", "round",
> > +								   the_repository, "round-%d",
> > +								   negotiation_round);
> >  					flushes = 0;
> >  					multi_ack = 0;
> >  					retval = 0;
> > @@ -490,6 +504,7 @@ static int find_common(struct fetch_negotiator *negotiator,
> >  						const char *hex = oid_to_hex(result_oid);
> >  						packet_buf_write(&req_buf, "have %s\n", hex);
> >  						state_len = req_buf.len;
> > +						haves++;
> >  						/*
> >  						 * Reset in_vain because an ack
> >  						 * for this commit has not been
> > @@ -510,14 +525,26 @@ static int find_common(struct fetch_negotiator *negotiator,
> >  			flushes--;
> >  			if (got_continue && MAX_IN_VAIN < in_vain) {
> >  				print_verbose(args, _("giving up"));
> > +				trace2_region_leave_printf("negotiation_v0_v1", "round",
> > +							   the_repository, "round-%d",
> > +							   negotiation_round);
> >  				break; /* give up */
> >  			}
> > -			if (got_ready)
> > +			if (got_ready) {
> > +				trace2_region_leave_printf("negotiation_v0_v1", "round",
> > +							   the_repository, "round-%d",
> > +							   negotiation_round);
> >  				break;
> > +			}
> > +			trace2_region_leave_printf("negotiation_v0_v1", "round",
> > +						   the_repository, "round-%d",
> > +						   negotiation_round);
> >  		}
> 
> Having many duplicated calls to "leave" makes the whole thing look
> somewhat confused.  Is this primarily because we have too many
> "break" that breaks out of the loop?

Yes, although now that I look at it again, I believe as soon as we
finish the do-while, we know that the round is over, so I believe we can
just put a single region_leave there. Fixed in V2.


> > @@ -1603,6 +1632,7 @@ static struct ref *do_fetch_pack_v2(struct fetch_pack_args *args,
> >  	struct oidset common = OIDSET_INIT;
> >  	struct packet_reader reader;
> >  	int in_vain = 0, negotiation_started = 0;
> > +	int negotiation_round = 0;
> >  	int haves_to_send = INITIAL_FLUSH;
> >  	struct fetch_negotiator negotiator_alloc;
> >  	struct fetch_negotiator *negotiator;
> > @@ -1659,6 +1689,10 @@ static struct ref *do_fetch_pack_v2(struct fetch_pack_args *args,
> >  						    "negotiation_v2",
> >  						    the_repository);
> >  			}
> > +			negotiation_round++;
> > +			trace2_region_enter_printf("negotiation_v2", "round",
> > +						   the_repository, "round-%d",
> > +						   negotiation_round);
> >  			if (send_fetch_request(negotiator, fd[1], args, ref,
> >  					       &common,
> >  					       &haves_to_send, &in_vain,
> > @@ -1686,12 +1720,20 @@ static struct ref *do_fetch_pack_v2(struct fetch_pack_args *args,
> >  			} else {
> >  				do_check_stateless_delimiter(args->stateless_rpc, &reader);
> >  				state = FETCH_SEND_REQUEST;
> > +				trace2_region_leave_printf("negotiation_v2", "round",
> > +							   the_repository, "round-%d",
> > +							   negotiation_round);
> >  			}
> >  			break;
> >  		case FETCH_GET_PACK:
> > +			trace2_region_leave_printf("negotiation_v2", "round",
> > +						   the_repository, "round-%d",
> > +						   negotiation_round);
> 
> Hmph.  Doesn't this logically belong to the "FETCH_PROCESS_ACKS"
> case arm?  We "leave" the current round when we did not get "ready"
> and "enter" the next round by moving to "FETCH_SEND_REQUEST" state,
> but at the same location when we did get "ready", we can "leave" the
> current (and final) round and move to "FETCH_GET_PACK" state.  I
> suspect the code structure and control flow becomes easier to see
> when expressed that way.

That works, but then we also need an additional region_leave in the
FETCH_SEND_REQUEST case where we jump directly to FETCH_GET_PACK. I'm
not sure which way is more intuitive, but I've gone with your suggestion
for V2.


> >  			trace2_region_leave("fetch-pack",
> >  					    "negotiation_v2",
> >  					    the_repository);
> > +			trace2_data_intmax("negotiation_v2", the_repository,
> > +					   "total_rounds", negotiation_round);
> >  			/* Check for shallow-info section */
> >  			if (process_section_header(&reader, "shallow-info", 1))
> >  				receive_shallow_info(args, &reader, shallows, si);