Use the system insights dashboards and built-in statistics tables to identify transactions that cause high latencies in Spanner Omni. To ensure the consistency of multiple concurrent transactions, Spanner Omni uses locks to control access to the data. Lock contention occurs when many transactions require frequent access to the same lock, leading to high latencies.
Spanner Omni operations acquire locks when the operations are part of a read-write transaction. Read-only transactions don't acquire locks.
To identify transactions that cause high latencies, follow these steps:
Before you begin
If you haven't already, download and install the Spanner Omni console. For more information see Spanner Omni downloads and Start the Spanner Omni console.
Check for a spike in latencies using system insights
Spanner Omni doesn't use Cloud Monitoring. Instead, use the system insights dashboard in the Spanner Omni console or Grafana. To check for latency spikes in the Spanner Omni console, do the following:
In the Spanner Omni console, click System Insights in the navigation pane.
In the system insights dashboard, check the latency charts for the following:
Request latency (
P50,P90,P99)Transaction latency (
P50,P90,P99)
Check the 99th percentile (
P99) for write operations on the latency charts. If you observe a spike in latency without a corresponding spike in CPU utilization or errors, the latency is likely due to lock contention issues.
Check for lock contention issues
To check if high latencies are due to lock contention, use the lock wait time metric available in the System Insights dashboard.
Check for high lock wait time
Check for high lock wait time in the Spanner Omni console system insights dashboard:
Locate the Lock wait time chart, which shows the total lock wait time for lock conflicts for the selected database in a 5-minute interval.
Check if this metric shows an increase that correlates with the latency spike you observed.
Analyze the lock wait data using system tables
After you confirm that lock contention is the cause of high latencies, use system statistics tables to analyze lock wait data and identify the transactions that are causing the contention. You can use the Spanner Omni CLI to query statistics tables.
For details on lock statistics tables, see Lock statistics in the Spanner documentation.
Identify contending transactions
To pinpoint the specific transactions that are contributing to high latencies, examine the transaction statistics for your database. Focus on transactions with high average latency. Optimize the transaction shape to reduce latencies. Consider applying the recommended practices to reduce lock contention in the Spanner documentation.
For details on transaction statistics tables, see Transaction statistics in the Spanner documentation.