Graph Metrics¶

Introduction¶

Currently, Nebula Graph supports obtaining the basic performance metric for the graph service via HTTP.

Each performance metrics consists of three parts, namely <counter_name>.<statistic_type>.<time_range>.

Counter Names¶

Each counter name is composed of the interface name and the counter name. Currently, the supported interfaces are:

graph_storageClient // Requests sent via storageClient, when sending requests to multiple storages concurrently, counted as one
graph_metaClient // Requests sent via metaClient
graph_graph_all // Requests sent by the client to the graph, when a request contains multiple queries, counted as one
graph_insertVertex // Insert a vertex
graph_insertEdge // Insert an edge
graph_deleteVertex // Delete a vertex
graph_deleteEdge // Delete an edge // Not supported yet
graph_updateVertex // Update properties of a vertex
graph_updateEdge // Update properties of an edge
graph_go // Execute the go command
graph_findPath // Find the shortest path or the full path
graph_fetchVertex // Fetch the vertex's properties. Only count the commands executed rather than the total number of fetched vertices.
graph_fetchEdge // Fetch the edge's properties. Only count the commands executed rather than the total number of fetched edges.

Each interface has three metrics, namely latency (in the units of us), QPS and QPS with errors. The suffixes are as follows:

_latency
_qps
_error_qps

The complete metric concatenates the interface name with the corresponding metric, such as graph_insertVertex_latency, graph_insertVertex_qps and graph_insertVertex_error_qps, representing the latency of inserting a vertex, QPS and the QPS with errors, respectively.

Statistics Type¶

Currently supported types are SUM, COUNT, AVG, RATE and P quantiles (P99, P999, ..., P999999). Among which:

Metrics have suffixes _qps and _error_qps support SUM, COUNT, AVG, RATE but don't support P quantiles.
Metrics have suffixes _latency support SUM, COUNT, AVG, RATE, and P quantiles.

Time Range¶

Currently, the supported time ranges are 60s, 600s, and 3600s, which correspond to the last minute, the last ten minutes, and the last hour till now.

Obtain the Corresponding Metrics via HTTP Interface¶

According to the above introduction, you can make a complete metrics name. Here are some examples:

graph_insertVertex_latency.avg.60   // the average latency of successfully inserting a vertex in the last minute
graph_updateEdge_error_qps.count.3600  // total number of failures in updating an edge in the last hour

Assume that a graph service is started locally, and the ws_http_port port number is set to 13000 when starting. It is sent through the GET interface of HTTP. The method name is get_stats, and the parameter is stats plus the corresponding metrics name. Here's an example of getting metrics via the HTTP interface:

# obtain a metrics
curl -G "http://127.0.0.1:13000/get_stats?stats=graph_insertVertex_qps.rate.60"
# graph_insertVertex_qps.rate.60=3069

# obtain multiple metrics at the same time
curl -G "http://127.0.0.1:13000/get_stats?stats=graph_insertVertex_qps.rate.60, graph_deleteVertex_latency.avg.60"
# graph_insertVertex_qps.rate.60=3069
# graph_deleteVertex_latency.avg.60=837

# obtain multiple metrics at the same time and return in json format
curl -G "http://127.0.0.1:13000/get_stats?stats=graph_insertVertex_qps.rate.60, graph_deleteVertex_latency.avg.60&returnjson"
# [{"value":2373,"name":"graph_insertVertex_qps.rate.60"},{"value":760,"name":"graph_deleteVertex_latency.avg.60"}]

# obtain all the metrics
curl -G "http://127.0.0.1:13000/get_stats?stats"
# or
curl -G "http://127.0.0.1:13000/get_stats"

Last update: April 8, 2021