HNSW Index creation fails on Greenplum (PostgreSQL 12.12)

[ahmedrachid]

Hello, I'm trying to install pg_embedding extension on Greenplum 7 (PostgreSQL 12.12 MPP), but getting the following errors when running: make PG_CONFIG=$GPHOME/bin/pg_config install (fyi: pgvector works fine with Greenplum)

gcc -Wall -Wmissing-prototypes -Wpointer-arith -Werror=vla -Wendif-labels -Wmissing-format-attribute -Wformat-security -fno-strict-aliasing -fwrapv -fexcess-precision=standard -Wno-unused-but-set-variable -Werror=implicit-fallthrough=3 -Wno-format-truncation -Wno-stringop-truncation -O3 -fargument-noalias-global -fno-omit-frame-pointer -g  -Werror=uninitialized -Werror=implicit-function-declaration -O3 -fPIC -I. -I./ -I/usr/local/greenplum-db-7.0.0-beta.3/include/postgresql/server -I/usr/local/greenplum-db-7.0.0-beta.3/include/postgresql/internal   -D_GNU_SOURCE -I/usr/include/libxml2   -c -o embedding.o embedding.c
embedding.c: In function ‘hnsw_options’:
embedding.c:401:19: error: implicit declaration of function ‘build_reloptions’; did you mean ‘view_reloptions’? [-Werror=implicit-function-declaration]
  return (bytea *) build_reloptions(reloptions, validate,
                   ^~~~~~~~~~~~~~~~
                   view_reloptions
embedding.c:401:9: warning: cast to pointer from integer of different size [-Wint-to-pointer-cast]
  return (bytea *) build_reloptions(reloptions, validate,
         ^
embedding.c: In function ‘hnsw_handler’:
embedding.c:513:13: error: ‘IndexAmRoutine’ {aka ‘struct IndexAmRoutine’} has no member named ‘amoptsprocnum’; did you mean ‘amoptions’?
  amroutine->amoptsprocnum = 0;
             ^~~~~~~~~~~~~
             amoptions
embedding.c:527:11: error: ‘IndexAmRoutine’ {aka ‘struct IndexAmRoutine’} has no member named ‘amusemaintenanceworkmem’
  amroutine->amusemaintenanceworkmem = false; /* not used during VACUUM */
           ^~
embedding.c:528:11: error: ‘IndexAmRoutine’ {aka ‘struct IndexAmRoutine’} has no member named ‘amparallelvacuumoptions’
  amroutine->amparallelvacuumoptions = VACUUM_OPTION_PARALLEL_BULKDEL;
           ^~
embedding.c:528:39: error: ‘VACUUM_OPTION_PARALLEL_BULKDEL’ undeclared (first use in this function); did you mean ‘CURSOR_OPT_PARALLEL_OK’?
  amroutine->amparallelvacuumoptions = VACUUM_OPTION_PARALLEL_BULKDEL;
                                       ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
                                       CURSOR_OPT_PARALLEL_OK
embedding.c:528:39: note: each undeclared identifier is reported only once for each function it appears in
embedding.c:534:22: warning: assignment to ‘aminsert_function’ {aka ‘_Bool (*)(struct RelationData *, long unsigned int *, _Bool *, struct ItemPointerData *, struct RelationData *, enum IndexUniqueCheck,  struct IndexInfo *)’} from incompatible pointer type ‘_Bool (*)(struct RelationData *, Datum *, _Bool *, ItemPointerData *, struct RelationData *, IndexUniqueCheck,  _Bool,  IndexInfo *)’ {aka ‘_Bool (*)(struct RelationData *, long unsigned int *, _Bool *, struct ItemPointerData *, struct RelationData *, enum IndexUniqueCheck,  _Bool,  struct IndexInfo *)’} [-Wincompatible-pointer-types]
  amroutine->aminsert = hnsw_insert;
                      ^
embedding.c:543:11: error: ‘IndexAmRoutine’ {aka ‘struct IndexAmRoutine’} has no member named ‘amadjustmembers’
  amroutine->amadjustmembers = NULL;
           ^~
embedding.c: In function ‘l2_distance’:
embedding.c:583: warning: ignoring #pragma clang loop [-Wunknown-pragmas]
  #pragma clang loop vectorize(enable)```

[ahmedrachid]

Hello, just updating / renaming this Github issue to "HNSW Index creation fails on Greenplum (PostgreSQL 12.12)"

Here's what I'm getting as error:

warehouse=# CREATE INDEX ON t USING hnsw (val) WITH (maxelements = 10, dims=3, m=3);
ERROR:  could not map shared memory segment "/PostgreSQL.90204": Invalid argument (dsm_impl.c:327)  (seg13 10.128.0.20:8001 pid=260965) (dsm_impl.c:327)
warehouse=#

[bayandin]

Thanks for the report!

The code has changed a lot since the bug was reported (see https://neon.tech/blog/pg_embedding-on-disk-hnsw-index).
Could you please check if it's still reproducible (you can use this PR where's the compilation for Postgres 12 is fixed).