From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Date: Mon, 28 Jan 2019 15:15:40 +0300 From: Vladimir Davydov Subject: Re: [tarantool-patches] [PATCH v2 4/5] Emit NOP if an applier skips row Message-ID: <20190128121540.gppxu7f544wgpi67@esperanza> References: <98c96d540bd0d23cdbdf42bb0d91c699afdd3e70.1548152776.git.georgy@tarantool.org> MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Disposition: inline In-Reply-To: <98c96d540bd0d23cdbdf42bb0d91c699afdd3e70.1548152776.git.georgy@tarantool.org> To: Georgy Kirichenko Cc: tarantool-patches@freelists.org List-ID: On Tue, Jan 22, 2019 at 01:31:12PM +0300, Georgy Kirichenko wrote: > Fill lsn gaps with NOP rows if applier configured to skip conflicting > rows. This enforces wal consistency. The patch may be worthwhile, but the comment sounds very obscure. What's "wal consistency"? How can the wal become inconsistent without this patch? Shouldn't there be a test proving this patch is correct? > > Prerequisite #980 > --- > src/box/applier.cc | 29 ++++++++++++++++------------- > 1 file changed, 16 insertions(+), 13 deletions(-) > > diff --git a/src/box/applier.cc b/src/box/applier.cc > index 148c8ce5a..adbe88679 100644 > --- a/src/box/applier.cc > +++ b/src/box/applier.cc > @@ -533,23 +533,26 @@ applier_subscribe(struct applier *applier) > row.replica_id); > vclock_follow_xrow(&replicaset.applier.vclock, &row); > int res = xstream_write(applier->subscribe_stream, &row); > - if (res != 0) { > - struct error *e = diag_last_error(diag_get()); > + struct error *e = diag_last_error(diag_get()); > + if (res != 0 && e->type == &type_ClientError && > + box_error_code(e) == ER_TUPLE_FOUND && > + replication_skip_conflict) { > /** > * Silently skip ER_TUPLE_FOUND error if such > * option is set in config. > */ > - if (e->type == &type_ClientError && > - box_error_code(e) == ER_TUPLE_FOUND && > - replication_skip_conflict) > - diag_clear(diag_get()); > - else { > - /* Rollback lsn to have a chance for a retry. */ > - vclock_set(&replicaset.applier.vclock, > - row.replica_id, old_lsn); > - latch_unlock(latch); > - diag_raise(); > - } > + diag_clear(diag_get()); > + row.type = IPROTO_NOP; > + row.bodycnt = 0; > + res = xstream_write(applier->subscribe_stream, > + &row); A comment explaining why this is done would be nice to have. > + } > + if (res != 0) { > + /* Rollback lsn to have a chance for a retry. */ > + vclock_set(&replicaset.applier.vclock, > + row.replica_id, old_lsn); > + latch_unlock(latch); > + diag_raise(); > } > } > done: