GRAPH.MEMORY

The GRAPH.MEMORY command returns detailed memory consumption statistics for a specific graph in megabytes (MB). It provides insight into how much memory is used by various internal data structures such as nodes, edges, schemas, indices, and matrix representations. This command can be used to monitor memory consumption at the graph level, making it especially useful for debugging, monitoring, performance optimization, and capacity planning in FalkorDB deployments.

Syntax

GRAPH.MEMORY USAGE <graph-name> [SAMPLES <count>]

Arguments

Argument	Description
<graph-name>	The name of the graph to inspect.
SAMPLES <n>	(Optional) Number of samples to take when estimating memory usage. A higher number improves accuracy but increases computation time. The samples are averaged to estimate the total size. Defaults to 100 if not specified. The value is clamped to the range [1, 10000].

Return

The command returns a set of key-value pairs, where each pair represents a specific memory metric and its value (in MB), corresponding to different components of the graph. In RESP2, this is encoded as a flat array of alternating field names and values:

Metric Name / Field	Type	Description
total_graph_sz_mb	integer	Total memory consumed by the graph.
label_matrices_sz_mb	integer	Amount of memory used by label matrices (node labels tracking).
relation_matrices_sz_mb	integer	Amount of memory used by relationship type matrices (graph topology tracking).
amortized_node_block_sz_mb	integer	Memory used by node blocks (amortized node storage).
amortized_node_attributes_by_label_sz_mb	map	Memory used by node attributes, broken down by label. Each key is a label name and its value is the memory in MB.
amortized_unlabeled_nodes_attributes_sz_mb	integer	Memory used by attributes of nodes with no label.
amortized_edge_block_sz_mb	integer	Memory used by edge blocks (amortized edge storage).
amortized_edge_attributes_by_type_sz_mb	map	Memory used by edge attributes, broken down by relationship type. Each key is a relationship type name and its value is the memory in MB.
indices_sz_mb	integer	Amount of memory consumed by indices (if any).

Examples

Basic Usage

javascript shell

import { FalkorDB } from 'falkordb';
const db = await FalkorDB.connect({
  socket: { host: 'localhost', port: 6379 }
});
const graph = db.selectGraph('myGraph');
const memoryInfo = await graph.memoryUsage();
console.log(memoryInfo);

GRAPH.MEMORY USAGE myGraph

With Sampling

javascript shell

const memoryInfo = await graph.memoryUsage({ samples: 500 });
console.log(memoryInfo);

GRAPH.MEMORY USAGE myGraph SAMPLES 500

Sample Output

127.0.0.1:6379> GRAPH.MEMORY USAGE flights
 1) "total_graph_sz_mb"
 2) (integer) 1086
 3) "label_matrices_sz_mb"
 4) (integer) 96
 5) "relation_matrices_sz_mb"
 6) (integer) 64
 7) "amortized_node_block_sz_mb"
 8) (integer) 120
 9) "amortized_node_attributes_by_label_sz_mb"
10) 1) "Airport"
    2) (integer) 35
    3) "City"
    4) (integer) 12
11) "amortized_unlabeled_nodes_attributes_sz_mb"
12) (integer) 0
13) "amortized_edge_block_sz_mb"
14) (integer) 54
15) "amortized_edge_attributes_by_type_sz_mb"
16) 1) "ROUTE"
    2) (integer) 68
17) "indices_sz_mb"
18) (integer) 752

Frequently Asked Questions 5

What units does GRAPH.MEMORY report in?

All memory values are reported in megabytes (MB).

What does the SAMPLES parameter do?

The SAMPLES parameter controls how many samples are taken when estimating memory usage. Higher values improve accuracy but increase computation time. The default is 100, and valid range is 1 to 10000.

What is the most useful metric for capacity planning?

The total_graph_sz_mb metric gives the total memory consumed by the graph. For more detailed analysis, examine per-label and per-relationship-type breakdowns to identify which parts of your graph use the most memory.

Why might indices_sz_mb be large relative to the graph?

Indexes store additional data structures for fast lookups. If you have many indexes or indexes on high-cardinality properties, the index memory can be significant. Only create indexes that are actually used by your queries.

Can I run GRAPH.MEMORY on a production system?

Yes, but be aware that higher SAMPLES values increase computation time. Use a moderate sample count for routine monitoring and higher counts only for detailed analysis during maintenance windows.