General tips for using Compute Engine

This page describes tips that you might find helpful if you run into problems using Compute Engine.

For help troubleshooting specific issues, see one of the following sections:

Viewing different response formats

The Google Cloud CLI performs most of its actions by making REST API calls. The pretty-printed results show only the most important information returned by any specific command. To see the different response formats, use the --format flag which displays the response in different output formats, including json, yaml, and text. For example, to see a list of instances in JSON, use --format json:

gcloud compute instances list --format json

Viewing gcloud compute logs

The gcloud CLI creates and stores logs in a log file that you can query, located at $HOME/.config/gcloud/logs. To see the latest log file on a Linux-based operating system, run:

$ less $(find ~/.config/gcloud/logs | sort | tail -n 1)

The log file includes information about all requests and responses made using the gcloud CLI tool.

To automatically purge the log files created by the gcloud CLI, use the max_log_days property, which sets the maximum number of days to retain log files before deleting. The default setting is 30 days. If you set this property value to 0, it turns off log garbage collection and does not delete log files.

 gcloud config set core/max_log_days DAYS_TO_RETAIN_LOGS

Disable gcloud CLI file logging:

The file $HOME/.config/gcloud/logs consumes space on the local file system. The amount of logs generated might overwhelm the amount of space on the local file system, which can cause issues like:

  • Space utilization reaching 100% on the instance.
  • Failure to execute gcloud CLI logging commands because there is no space left to create a new file on the local file system.

To change the gcloud CLI behavior and disable the file logging, use the disable_file_logging property:

 gcloud config set core/disable_file_logging True

Selecting resource names

When selecting names for your resources, keep in mind that these friendly-names may be visible on support and operational dashboards within Compute Engine. For this reason, it is recommended that resource names that do not expose any sensitive information.

Communicating to the internet

An instance has direct internet access only if both of the following conditions are true:

Instances can also access the internet indirectly, by connecting through Cloud NAT or an instance-based proxy. For additional considerations, including firewall rule configuration, see Internet access requirements.

Idle connections

Google Cloud network components don't keep idle connections open indefinitely. To prevent dropped connections, consider how the following components handle idle connections:

To prevent idle TCP connections from being dropped, you can configure TCP keepalive, which keeps connections active by sending periodic packets (probes) to reset the idle timeouts:

  • Make sure that client or server applications create sockets with the SO_KEEPALIVE socket option. Consult your application or software library's documentation to determine how to open sockets with the SO_KEEPALIVE option.

  • Configure the following TCP keepalive parameters:

    • The inactivity period, representing the time that must elapse between the last non-keepalive packet and the first keepalive probe in a sequence.

    • The keepalive interval, representing the time between each TCP keepalive probe.

    • The keepalive probes, representing the total number of unacknowledged TCP keepalive probes that are sent in a sequence before the connection is considered dropped.

Client or server applications can set the TCP keepalive parameters by using socket options such as TCP_KEEPIDLE, TCP_KEEPINTVL, and TCP_KEEPCNT, or you can set the TCP keepalive parameters for your operating system.

The following examples show how to set an inactivity period of 60 seconds for your operating system:

Linux


Run the following command:

$ sudo /sbin/sysctl -w net.ipv4.tcp_keepalive_time=60 net.ipv4.tcp_keepalive_intvl=60 net.ipv4.tcp_keepalive_probes=5

To ensure that the settings survive a reboot, add the settings to your /etc/sysctl.conf file. These kernel parameters apply to both IPv4 and IPv6, even though they have ipv4 in their names.

See Linux TCP Keepalive HOWTO for additional information.

macOS


Run the following command:

$ sudo sysctl -w net.inet.tcp.always_keepalive=1 net.inet.tcp.keepidle=60000 net.inet.tcp.keepinit=60000 net.inet.tcp.keepintvl=60000

Windows


Under the registry path HKEY_LOCAL_MACHINE\SYSTEM\CurrentControlSet\Services\Tcpip\Parameters\, add the following settings, using the DWORD data type, or edit the values if the settings already exist:

KeepAliveInterval: 1000
KeepAliveTime: 60000
TcpMaxDataRetransmissions: 10

Accessing Compute Engine as a different SSH user

By default, the gcloud compute command-line tool uses the $USER variable to add users to the /etc/passwd file for connecting to instances using SSH. You can specify a different user using the --ssh-key-file PRIVATE_KEY_FILE flag when running the gcloud compute ssh command. For example:

gcloud compute ssh example-instance --ssh-key-file my-private-key-file

For more information, see the gcloud reference documentation.

Interacting with the serial console

You can enable interactive access to an instance's serial console so you can connect and troubleshoot instances through the serial console.

For more information, see Troubleshooting using the serial console.

Avoiding packet fragmentation to instances built from custom images

The VPC network has a default maximum transmission unit (MTU) of 1460 bytes for Linux images and Windows Server images. However, the network MTU can be changed. For details, see the maximum transmission unit overview in the VPC documentation.

When creating client applications that communicate with Compute Engine instances over UDP sockets, you can avoid fragmentation if you set the maximum size of the UDP datagram's data to 28 bytes less than the network MTU. For example, if the network's MTU is 1460 bytes, you can send up to 1432 bytes of UDP data per packet without fragmentation. If the network's MTU is 1500 bytes, you can send up to 1472 bytes of UDP data without fragmentation. The 28 bytes are used for an IPv4 packet header (20 bytes) and a UDP datagram header (8 bytes). You can set the network's MTU to a maximum of 8896 bytes.

Performance and CPU diagnostics

Unexpected latency spikes or crashes in your application on modern CPU platforms might indicate CPU bus locks. These issues occur when atomic operations are performed on unaligned memory.

To identify the issue, review your serial port output for the following log entry: x86/split lock detection: #DB: <process_name>/<pid> took a bus_lock trap at address: 0x<address>.

Serial console output enables the detection of hardware-level events, such as CPU bus lock traps, which indicate unaligned memory operations that can degrade system performance.

For more information, see Troubleshoot CPU bus locks.