Client-side metrics descriptions
Client-side metrics are available when you use the Bigtable client library for Java or the Bigtable HBase client for Java. This page provides a detailed description of each client-side metric.
To learn how to set up and configure client-side metrics, see Set up client-side metrics. For an overview, see the Client-side metrics overview.
Metrics are published to the region of the Bigtable cluster that
they are for. If Bigtable can't resolve the cluster for a metric,
like when an operation fails, then the metric is published to the global region
and the cluster
tag has the value unspecified
.
Use client-side metrics in conjunction with server-side metrics for performance debugging. For a comprehensive list of Bigtable server-side and client-side metrics, see Metrics.
Operation latencies
Attribute | Description |
---|---|
Full metric | bigtable.googleapis.com/client/operation_latencies |
Definition | The total end-to-end latency across all RPC attempts associated with
a Bigtable operation. This metric measures an
operation's round trip from the client to Bigtable and
back to the client and includes all retries. For ReadRows requests, the operation latencies include
the application processing time for each returned message. |
Supported Bigtable operations |
|
Type | Distribution |
Available dimensions |
|
Attempt latencies
Attribute | Description |
---|---|
Full metric | bigtable.googleapis.com/client/attempt_latencies |
Definition | The latencies of a client RPC attempt. Under normal circumstances, this value is identical to operation_latencies . If the client receives transient
errors, however, then operation_latencies is the sum of
all attempt_latencies and the exponential delays. |
Supported Bigtable operations |
|
Type | Distribution |
Available dimensions |
|
Retry count
Attribute | Description |
---|---|
Full metric | bigtable.googleapis.com/client/retry_count |
Definition | A counter that records the number of attempts that an operation required to complete. Under normal circumstances, this value is empty. |
Supported Bigtable operations |
|
Type | INT64 |
Available dimensions |
|
Application blocking latencies
Attribute | Description |
---|---|
Full metric | bigtable.googleapis.com/client/application_blocking_latencies |
Definition | The time from when the client receives the response to a
request until the application reads the response. This metric is most
relevant for ReadRows requests. The start and stop times
for this metric depend on the way that you send the read request; see
Application blocking latencies timer
examples for details.
|
Supported Bigtable operations |
|
Type | Distribution |
Available dimensions |
|
Application blocking latencies timer examples
ReadRows
requests are typically sent in one of three ways: iteratively,
asynchronously, or as a future call. The following examples demonstrate the
start and stop times for the application blocking latencies for each type of
ReadRows
operation.
Iterative
long totaLatency=0;
ServerStream<Row> stream = bigtableDataClient.readRows(query);
for (Row row : stream) {
timer.start();
// Do something with row
timer.stop();
totalLatency += timer.elapsed();
}
Asynchronous
client.readRowsAsync(query, new ResponseObserver<Row>() {
@Override
public void onStart(StreamController streamController) {
}
@Override
public void onResponse(Row row) {
timer.start();
// Do something with the row
timer.stop();
totalLatency += timer.elapsed();
}
@Override
public void onError(Throwable throwable) {
// Do something with the error
}
@Override
public void onComplete() {
}
});
Future call
The application blocking latencies should be approximately 0 (no timer) for
this type of ReadRows
request.
List<Row> rows = bigtableDataClient.readRowsCallable().all().call(query);
First-response latencies
Attribute | Description |
---|---|
Full metric | bigtable.googleapis.com/client/first_response_latencies |
Definition | Latencies from when a client sends a request and receives the first row of the response. |
Supported Bigtable operation | ReadRows |
Type | Distribution |
Available dimensions |
|
Server latencies
Attribute | Description |
---|---|
Full metric | bigtable.googleapis.com/client/server_latencies |
Definition | Latencies between the time when the Google frontend receives an RPC and when it sends the first byte of the response. |
Supported Bigtable operations |
|
Type | Distribution |
Available dimensions |
|
Connectivity error count
Attribute | Description |
---|---|
Full metric | bigtable.googleapis.com/client/connectivity_error_count |
Definition | The number of requests that failed to reach Google's network. In normal cases, this number is 0. When the number is not 0, it can indicate connectivity issues between the application and the Google network. |
Supported Bigtable operations |
|
Type | Distribution |
Available dimensions |
|
Client blocking latencies
Attribute | Description |
---|---|
Full metric | bigtable.googleapis.com/client/client_blocking_latencies |
Definition | Latencies introduced when the client blocks the sending of more requests to the server because of too many pending requests in a bulk operation. For versions 2.21.0 and later, this metric also includes the latencies of requests queued on gRPC channels. |
Supported Bigtable operations |
|
Type | Distribution |
Available dimensions |
|