This topic describes the versions of the CloudMonitor agent.
3.5.12
Item | Description |
Release date | 2024-10-09 |
New feature | New regions are supported: Hangzhou KS01 ACDR China West 1 Antgroup
Deleted region: India (Mumbai) The clock for task scheduling is changed from system clock to monotonic clock to prevent timing confusion caused by system time adjustment. The tool curl command is added to the command line to detect a monitored address and print the interaction process. This facilitates troubleshooting.
/usr/local/cloudmonitor/bin/argusagent tool curl --help
Usage: argusagent tool curl [options] url
Allowed options:
-h [ --help ] Print this help message
-X [ --request ] arg (=GET) Specifies a custom request method to use.
--url arg Target url.
-d [ --data ] arg Only for POST, http bod.y
-H [ --header ] arg Extra header to use.
-m [ --max-time ] arg (=30) Maximum time in seconds that you allow the whole operation to take.
-x [ --proxy ] arg Use the specified proxy, format: [protocol://]host[:port].
--proxy-user arg Specify the user name to use for proxy authentication.
--proxy-pass arg Specify the password to use for proxy authentication.
--proxy-http2 Negotiate HTTP version 2 with an HTTPS proxy. The proxy might still only offer HTTP/1 and
then curl sticks to using that version. This has no effect for any other kinds of
proxies.
--json arg Json object config, this will ignore all other options.
--json-file arg Json object config file, this will ignore all other options.
--task-id arg Detect once of http task with taskId
The -e GetTopTasks parameter is added to the command line so that you can dynamically view the time consumption of the top 20 tasks during runtime.
# The path on CoreOS is /opt/cloudmonitor/bin/argusagent.
/usr/local/cloudmonitor/bin/argusagent -e GetTopTasks
"C:\Program Files\Alibaba\cloudmonitor\bin\argusagent.exe -e GetTopTasks
Proxy information is added to the log during heartbeats and metric reporting. This prevents the misunderstanding that the agent is using a public network. Process monitoring supports labeling. A self-monitoring metric is added: If no basic metrics are collected for 2 consecutive minutes, the system automatically restarts. GPU: Ascend and Hygon GPUs are supported. These GPUs apply only to Linux. The agent installation script is upgraded to 1.13. The dependency on bash is accurate to sh, which further improves the adaptability of scripts and supports a wider range of installation and operation, such as Android. User-defined proxies are supported for manual installation. Previously, you can only use a custom proxy to download the agent installation script. A self-check is performed after the installation package is downloaded. The old version is uninstalled only if the installation package is valid. This prevents the issue that a wget package is downloaded successfully but an empty package is actually downloaded. The installation log is optimized so that the log is more readable and localizable. Windows supports installation based on the local installation package (-packageFile), skipping the process of downloading the installation package. The following bug that occurs when you use a proxy to install the CloudMonitor agent on a third-party host is fixed: After an installation package is successfully downloaded, the package is downloaded again without using a proxy. Cloud Phone is supported.
|
Fixed issue | The argusagent service was not called as a service during the installation. Availability monitoring does not support Multi Headers. Multiple IP addresses are returned after the hostname -i command is run on Linux. When Telnet is created by calling API operations, Telnet supports the mixed use of hosts and Uniform Resource Identifiers (URIs). In non-ECS mode, the accesskey.properties file in the bin directory cannot be identified. On Linux, udevadm is repeatedly called when a disk does not have a serial number. HTTP/2 does not take effect when it is used as a proxy. If the packet loss rate of the ping task is large, task scheduling is not precisely determined. Due to unreasonable timing assumption of the three-timer algorithm, it has been reconstructed to be triggered by events (double events of receiving returned results and timeout). When an availability monitoring task is updated, the old task may occasionally fail to be cleared. The SIGSEGV issue that occasionally occurs is resolved when the availability monitoring task is updated. The following issue is fixed: The memory metrics fail to be obtained when Windows Perf data is abnormal or missing. Memory leaks on Windows. The value returned by CommandLineToArgvW uses GlobalFree but not LocalFree for memory release.
|
3.5.11
Item | Description |
Release date | 2024-03-25 |
New feature | Windows x64 is officially supported. IPv6 is supported. HTTP/2 is supported. macOS and FreeBSD (based on the Sigar library) are supported. Seven proxy protocols are supported: HTTP, HTTPS, HTTPS/2, SOCKS4, SOCKS4A, SOCKS5, and SOCKS5h. GPUs: C:\Windows\System32\nvidia-smi.exe is supported.
A GPU can be installed during the runtime of the CloudMonitor agent. The GPU can be installed after the agent is installed. GPU metric data can be collected more securely and faster based on the dynamic library (libnvml). Note You must manually enable the dynamic library by setting nvidia.nvml.enabled to true. This prevents the dynamic library from being stuck if it is enabled by running commands on some systems.
The effective time of availability detection supports cron expressions. Prometheus-based data collection supports HTTP header authentication. The following regions are supported: The following four metrics are removed: system.udp system.task: the number of system processes or threads memory.swap: the Linux-only swap partitions system.cpuCore: the CPU core metrics
|
Fixed issue | Only the issue that the process memory exceeds 4 GB is reported when the following two issues occur on Windows: Top 5 processes with the highest load are monitored and the process memory exceeds 4 GB. DNS resolution is stuck (DNS resolution takes more than 20 seconds on some systems). Some Prometheus metrics fail to be parsed. Log collection requires high CPU utilization. The availability monitoring does not continue the next time if the ping fails. The serial number of the host may contain a line feed. The same Telnet task for detection is run multiple times, which may cause ArgusAgent to crash. SOCKS5 support is not standardized. The Windows Management Instrumentation Command-line (WMIC) is not found on Windows. The CloudMonitor agent fails to be started because std::locale("") is not supported. A large number of memory leaks. The SIGSEGV issue caused by the localTimeCache failure when the main function exits. Note A core dump file is generated during a normal exit.
|
Performance optimization | The stability is improved. The maximum number of system processes monitored by CloudMonitor can be configured. For example, if you set the threshold to 5,000, process monitoring stops when the number of system processes reaches 5,000. This prevents excessive resource usage in multiple regions. When you upgrade and install the CloudMonitor agent, the installation package is automatically downloaded. When the installation is successful, the installation package is automatically deleted. The compatibility of JSON configuration files is improved to support C-style comments, allow commas at the end, and allow non-standard UTF-8 encoding. Log collection in public cloud: Non-JSON prefixes and suffixes are supported for JSON resolution. Timing for disk data collection is changed from the system clock to the hardware clock to prevent timing errors caused by system clock adjustment. The value of mount_point/dir_name for disk data collection is a string of concatenated directories of all disks mounted on dir_name. The maximum length is 2,048 bytes. Note The value can be adjusted by the agent.resource.dirName.limit parameter. The default length is 2,048 bytes. If the length is less than 1,024 bytes, it is set to 1,024 bytes.
|
3.5.10
Item | Description |
Release date | 2023-09-08 |
New feature | moduleTask.json supports feature disabling. You no longer need to disable a feature by deleting it.
The argusagent tool top tool is provided for you to sort and show the top N-n N from the following three dimensions: -by fd (number of opened files), -by mem (memory usage), and -by cpu (CPU utilization). GPU metric data collection can be disabled. GPU metric data can be dynamically collected during the runtime of the CloudMonitor agent. The number of CPU cores can be changed dynamically during the runtime of the CloudMonitor agent. Hosts that are not provided by Alibaba Cloud but managed by Cloud Assistant are supported.
Note The CloudMonitor agent generates a minidump file when it exits unexpectedly. When the CloudMonitor agent starts next time, the minidump file is reported to CloudMonitor for analysis and bug fixing in the background, improving the stability of the CloudMonitor agent. |
Fixed issue | Data failed to be reported in a specified time zone. The number of opened files for a process ranking top five is incorrect. When a process ID is greater than 1,000,000, the precision of the reported data is degraded. After the CloudMonitor agent is installed on the Windows hosts of some existing users, argusagent cannot be started or the specified dynamic-link library (DLL) is missing. A carriage return line feed (CRLF) injection occurs when a configuration file on Windows is read. When the number of processes is collected for the first time, the collection is skipped, causing the process quantity to be 0.
|
Performance optimization | Process monitoring performance is optimized to reduce the frequency of unexpected agent exits. The installation success rate of the CloudMonitor agent on Windows is increased. When the resource usage of the CloudMonitor agent is exceeded, the top 10 resources consumed (for example, CPU utilization) are collected and the call stacks of all threads are enumerated to help analyze the resource usage.
|
3.5.8
Item | Description |
Release date | 2022-06-30 |
New feature | Files can be distributed and stored as required. The following metrics are supported: the packet loss rate, the error rate, and the number of deadlock processes. The following metrics are supported: the device usage and the swap usage.
|
Fixed issue | The size of the dir_name field is reduced to 512 bytes when it is concatenated. On Windows, GetUptime is called when a system module is called. The IphlpapiGetTcpTable function causes memory leaks. The cpu.total metric is changed from cpuPercent.combined to 1-cpu.idle. The ping probe checks whether the destination IP address used when data is sent is the same as the source IP address used when data is received. ICMP serial number collision is also fixed. On Windows, a service fails to start due to spaces contained in the argusagent path (imagePath in the registry).
|
3.5.7
Item | Description |
Release date | 2022-04-30 |
New feature | TCP metrics are supported. |
Fixed issue | Excessive data is reported when disk directories are concatenated. The CloudMonitor agent does not perform polling on the international site (alibabacloud.com). Win32 changes the .py file associations.
|
3.5.5
Item | Description |
Release date | 2021-12-30 |
New feature | Log collection is supported. |
Fixed issue | None |
3.5.4
Item | Description |
Release date | 2021-12-16 |
New feature | |
Fixed issue | The number of CPU cores collected from Container Service is inaccurate in some cases. An error occurs during IP address resolution on Windows. The CloudMonitor agent occasionally exits when the process collects the number of threads. The ping command fails to return the result of availability tests. The scheduling interval of availability tests is larger than the configured time interval in some cases.
|
3.5.3
Item | Description |
Release date | 2021-09-10 |
New feature | Exporters can be used to collect data. Requests can be redirected to HTTPS in HTTP-based availability monitoring tasks. More SSL ciphers can be used in HTTP-based availability monitoring tasks. By default, HTTP-based availability monitoring tasks have the same performance as curl commands. Headers such as user_agent are added to the requests of HTTP-based availability monitoring tasks.
|
Fixed issue | The number of TCP connections is occasionally miscalculated on IPv6 hosts. The disk usage is precise to decimal places. The CloudMonitor agent is occasionally disabled when it collects the cred metric in process monitoring tasks.
|
3.5.2
Item | Description |
Release date | 2021-06-30 |
New feature | The monitoring frequency of availability monitoring tasks can be changed. The usability of availability monitoring tasks is optimized. For example, more comprehensive and standardized local logs are recorded. The Uptime metric can be used to view the time period from the last startup time of the system to the current time.
|
Fixed issue | The CloudMonitor agent fails to be installed on some Windows hosts due to the missing of DLLs. |
3.4.10
Item | Description |
Release date | 2021-03-11 |
New feature | None |
Fixed issue | The CloudMonitor agent fails to be installed on some hosts that are not provided by Alibaba Cloud. The installation fails because the path to the AccessKey pair cannot be read. |
3.4.9
Item | Description |
Release date | 2021-01-05 |
New feature | The SOCKS5 proxy is supported. |
Fixed issue | The CloudMonitor agent fails to be installed on some hosts that are not provided by Alibaba Cloud. The installation fails due to the missing of specific DLLs. These hosts run Windows Server 2012 or earlier. |
3.4.8
Item | Description |
Release date | 2020-11-17 |
New feature | None |
Fixed issue | URLs cannot be correctly parsed by availability monitoring. |
3.4.7
Item | Description |
Release date | 2020-07-27 |
New feature | More supported metrics: The CloudMonitor agent can collect metrics about disk I/O and the utilization of a single CPU core. Less resource usage: After optimization, the CloudMonitor agent occupies fewer resources and has less impact on the system performance. System protection: When the system load is heavy or the CloudMonitor agent occupies excessive system resources due to specific reasons, the CloudMonitor agent automatically exits.
|
Fixed issue | The CloudMonitor agent for Go automatically changes the system time. |
Elastic Compute Service (ECS)
Container Compute Service (ACS)
