Re: Logical replication and multimaster

Hi,

I have integrated pglogical_output in multimaster, using bdr_apply from BDR as template for implementation of receiver part.
The time of insert is reduced almost 10 times comparing with logical replication based on decoder_raw/receiver_raw plugins which performs logical replication using SQL statements. But unfortunately time of updates is almost not changed.
It is expected result because I didn't see any functions related with SQL parsing/preparing in profile.
Now in both cases profile is similar:
 
  4.62%  postgres                            [.] HeapTupleSatisfiesMVCC
  2.99%  postgres                            [.] heapgetpage
  2.10%  postgres                            [.] hash_search_with_hash_value
  1.86%  postgres                            [.] ExecProject
  1.80%  postgres                            [.] heap_getnext
  1.79%  postgres                            [.] PgXidInMVCCSnapshot

By the way, you asked about comments concerning pglogical_output. I have one: most of pglogical protocol functions have "PGLogicalOutputData *data" parameter. There are few exceptions:
   
    write_startup_message_fn, pglogical_write_origin_fn, pglogical_write_rel_fn

PGLogicalOutputData is the only way to pass protocol specific data, using "PGLogicalProtoAPI *api" field.
This field is assigned by  pglogical_init_api() function. And I can extend this PGLogicalProtoAPI structure by adding some protocol specific fields.
For example, this is how it is done now for multimaster:

typedef struct PGLogicalProtoMM
{
    PGLogicalProtoAPI api;
    bool isLocal; /* mark transaction as local */
} PGLogicalProtoMM;

PGLogicalProtoAPI *
pglogical_init_api(PGLogicalProtoType typ)
{
    PGLogicalProtoMM* pmm = palloc0(sizeof(PGLogicalProtoMM));
    PGLogicalProtoAPI* res = &pmm->api;
    pmm->isLocal = false;
    res->write_rel = pglogical_write_rel;
    res->write_begin = pglogical_write_begin;
    res->write_commit = pglogical_write_commit;
    res->write_insert = pglogical_write_insert;
    res->write_update = pglogical_write_update;
    res->write_delete = pglogical_write_delete;
    res->write_startup_message = write_startup_message;
    return res;
}

But I have to add "PGLogicalOutputData *data"  parameter to pglogical_write_rel_fn function.
Di you think that it will be better to pass this parameter to all functions?

May be it is not intended way of passing custom data to this functions...
Certainly it is possible to use static variables for this purpose.
But I think that passing user specific data through PGLogicalOutputData is safer and more flexible solution.



On 12/03/2015 04:53 PM, Craig Ringer wrote: