Skip to main content
Reindexing is necessary to change the index’s schema. This includes adding, removing, or renaming fields, or changing a field’s tokenizer configuration. The basic syntax for REINDEX is: 
REINDEX INDEX search_idx;
This operation takes an exclusive lock on the table, which blocks incoming writes (but not reads) while the new index is being built. To allow for concurrent writes during a reindex, use REINDEX CONCURRENTLY: 
REINDEX INDEX CONCURRENTLY search_idx;
The tradeoff is that REINDEX CONCURRENTLY is slower than a plain REINDEX. Generally speaking, REINDEX CONCURRENTLY is recommended for production systems that cannot tolerate temporarily blocked writes.
In order for REINDEX CONCURRENTLY to succeed, Postgres requires that the session that is executing the command remain open. If the session is closed, Postgres will cancel the reindex. This is relevant if you are using a connection pooler like pgbouncer, which can be configured to terminate sessions after a certain idle timeout is reached.
If REINDEX CONCURRENTLY fails or is cancelled, an invalid transient index will be left behind that must be dropped manually. To check for invalid indexes in psql, run \d <table_name> and look for indexes suffixed by _ccnew.