PostgreSQL Server Metrics

blks_hit: Number of times disk blocks were found already in the buffer cache, so that a physical disk read was not necessary. This only includes hits in the PostgreSQL buffer cache, does not include the operating system's file system .cache).

blks_read: Number of disk blocks read from the database.

confl_bufferpin: Number of queries in the database that were canceled because of pinned buffers.

confl_deadlock: Number of queries in the database that were canceled because of deadlocks

confl_lock: Number of queries in the database that were canceled due because of timeouts.

confl_snapshot: Number of queries in the database that were canceled because of old snapshots.

confl_tablespace: Number of queries in the database that were canceled because of dropped tablespaces.

numbackends: Number of backends currently connected to the database.

size_mb:

tup_deleted: Number of rows deleted by queries in the database.

tup_fetched: Number of rows fetched by queries in the database.

tup_inserted: Number of rows inserted by queries in the database.

tup_returned: Number of rows returned by queries in the database.

tup_updated: Number of rows updated by queries in the database.

xact_commit: Number of transactions in the database that have been committed.

xact_rollback: Number of transactions in the database that have been rolled back.

pgvector Metrics

Metrics Description
ART Measures the average time taken to execute Postgres queries, providing insights into query performance.
Queries per minute Records the average rate of queries executed per minute, reflecting the query load on the database.
Avg Docs Retrieved Represents the average number of documents or results retrieved.
Vectors Inserted The total number of vector embeddings inserted in the database.
Vector Insertion Rate Tracks the rate at which vector embeddings are inserted into the database.
Distance Metrics
Cosine distance frequency Counts the number of similarity searches performed using cosine distance, a metric for measuring angular similarity between vectors.
L1 distance frequency Tracks the usage of L1 distance (Manhattan distance) in searches, which calculates the sum of absolute differences between vector components.
L2 distance frequency Measures the frequency of L2 distance (Euclidean distance) searches, which computes the straight-line distance between vectors.
Inner Product frequency Measures the number of searches leveraging inner product, often used to measure the alignment or dot product similarity between vectors.
Hamming distance frequency Captures the frequency of searches using Hamming distance, a metric for comparing binary vectors by counting differing bit positions.
Jaccard distance frequency Tracks the frequency of searches using Jaccard distance, a metric for comparing binary vectors.