How to query logs from Azure Monitor for VMs (Preview)

Azure Monitor for VMs collects performance and connection metrics, computer and process inventory data, and health state information and forwards it to the Log Analytics data store in Azure Monitor. This data is available for search in Log Analytics. You can apply this data to scenarios that include migration planning, capacity analysis, discovery, and on-demand performance troubleshooting.

Map records

One record is generated per hour for each unique computer and process, in addition to the records that are generated when a process or computer starts or is on-boarded to Azure Monitor for VMs Map feature. These records have the properties in the following tables. The fields and values in the ServiceMapComputer_CL events map to fields of the Machine resource in the ServiceMap Azure Resource Manager API. The fields and values in the ServiceMapProcess_CL events map to the fields of the Process resource in the ServiceMap Azure Resource Manager API. The ResourceName_s field matches the name field in the corresponding Resource Manager resource.

There are internally generated properties you can use to identify unique processes and computers:

  • Computer: Use ResourceId or ResourceName_s to uniquely identify a computer within a Log Analytics workspace.
  • Process: Use ResourceId to uniquely identify a process within a Log Analytics workspace. ResourceName_s is unique within the context of the machine on which the process is running (MachineResourceName_s)

Because multiple records can exist for a specified process and computer in a specified time range, queries can return more than one record for the same computer or process. To include only the most recent record, add "| dedup ResourceId" to the query.

Connections

Connection metrics are written to a new table in Log Analytics - VMConnection. This table provides information about the connections for a machine (inbound and outbound). Connection Metrics are also exposed with APIs that provide the means to obtain a specific metric during a time window. TCP connections resulting from "accept-ing on a listening socket are inbound, while those created by connect-ing to a given IP and port are outbound. The direction of a connection is represented by the Direction property, which can be set to either inbound or outbound.

Records in these tables are generated from data reported by the Dependency agent. Every record represents an observation over a one-minute time interval. The TimeGenerated property indicates the start of the time interval. Each record contains information to identify the respective entity, that is, connection or port, as well as metrics associated with that entity. Currently, only network activity that occurs using TCP over IPv4 is reported.

To manage cost and complexity, connection records do not represent individual physical network connections. Multiple physical network connections are grouped into a logical connection, which is then reflected in the respective table. Meaning, records in VMConnection table represent a logical grouping and not the individual physical connections that are being observed. Physical network connection sharing the same value for the following attributes during a given one-minute interval, are aggregated into a single logical record in VMConnection.

Property Description
Direction Direction of the connection, value is inbound or outbound
Machine The computer FQDN
Process Identity of process or groups of processes, initiating/accepting the connection
SourceIp IP address of the source
DestinationIp IP address of the destination
DestinationPort Port number of the destination
Protocol Protocol used for the connection. Values is tcp.

To account for the impact of grouping, information about the number of grouped physical connections is provided in the following properties of the record:

Property Description
LinksEstablished The number of physical network connections that have been established during the reporting time window
LinksTerminated The number of physical network connections that have been terminated during the reporting time window
LinksFailed The number of physical network connections that have failed during the reporting time window. This information is currently available only for outbound connections.
LinksLive The number of physical network connections that were open at the end of the reporting time window

Metrics

In addition to connection count metrics, information about the volume of data sent and received on a given logical connection or network port are also included in the following properties of the record:

Property Description
BytesSent Total number of bytes that have been sent during the reporting time window
BytesReceived Total number of bytes that have been received during the reporting time window
Responses The number of responses observed during the reporting time window.
ResponseTimeMax The largest response time (milliseconds) observed during the reporting time window. If no value, the property is blank.
ResponseTimeMin The smallest response time (milliseconds) observed during the reporting time window. If no value, the property is blank.
ResponseTimeSum The sum of all response times (milliseconds) observed during the reporting time window. If no value, the property is blank.

The third type of data being reported is response time - how long does a caller spend waiting for a request sent over a connection to be processed and responded to by the remote endpoint. The response time reported is an estimation of the true response time of the underlying application protocol. It is computed using heuristics based on the observation of the flow of data between the source and destination end of a physical network connection. Conceptually, it is the difference between the time the last byte of a request leaves the sender, and the time when the last byte of the response arrives back to it. These two timestamps are used to delineate request and response events on a given physical connection. The difference between them represents the response time of a single request.

In this first release of this feature, our algorithm is an approximation that may work with varying degree of success depending on the actual application protocol used for a given network connection. For example, the current approach works well for request-response based protocols such as HTTP(S), but does not work with one-way or message queue-based protocols.

Here are some important points to consider:

  1. If a process accepts connections on the same IP address but over multiple network interfaces, a separate record for each interface will be reported.
  2. Records with wildcard IP will contain no activity. They are included to represent the fact that a port on the machine is open to inbound traffic.
  3. To reduce verbosity and data volume, records with wildcard IP will be omitted when there is a matching record (for the same process, port, and protocol) with a specific IP address. When a wildcard IP record is omitted, the IsWildcardBind record property with the specific IP address, will be set to "True" to indicate that the port is exposed over every interface of the reporting machine.
  4. Ports that are bound only on a specific interface have IsWildcardBind set to "False".

Naming and Classification

For convenience, the IP address of the remote end of a connection is included in the RemoteIp property. For inbound connections, RemoteIp is the same as SourceIp, while for outbound connections, it is the same as DestinationIp. The RemoteDnsCanonicalNames property represents the DNS canonical names reported by the machine for RemoteIp. The RemoteDnsQuestions and RemoteClassification properties are reserved for future use.

Geolocation

VMConnection also includes geolocation information for the remote end of each connection record in the following properties of the record:

Property Description
RemoteCountry The name of the country hosting RemoteIp. For example, United States
RemoteLatitude The geolocation latitude. For example, 47.68
RemoteLongitude The geolocation longitude. For example, -122.12

Malicious IP

Every RemoteIp property in VMConnection table is checked against a set of IPs with known malicious activity. If the RemoteIp is identified as malicious the following properties will be populated (they are empty, when the IP is not considered malicious) in the following properties of the record:

Property Description
MaliciousIp The RemoteIp address
IndicatorThreadType Threat indicator detected is one of the following values, Botnet, C2, CryptoMining, Darknet, DDos, MaliciousUrl, Malware, Phishing, Proxy, PUA, Watchlist.
Description Description of the observed threat.
TLPLevel Traffic Light Protocol (TLP) Level is one of the defined values, White, Green, Amber, Red.
Confidence Values are 0 – 100.
Severity Values are 0 – 5, where 5 is the most severe and 0 is not severe at all. Default value is 3.
FirstReportedDateTime The first time the provider reported the indicator.
LastReportedDateTime The last time the indicator was seen by Interflow.
IsActive Indicates indicators are deactivated with True or False value.
ReportReferenceLink Links to reports related to a given observable.
AdditionalInformation Provides additional information, if applicable, about the observed threat.

ServiceMapComputer_CL records

Records with a type of ServiceMapComputer_CL have inventory data for servers with the Dependency agent. These records have the properties in the following table:

Property Description
Type ServiceMapComputer_CL
SourceSystem OpsManager
ResourceId The unique identifier for a machine within the workspace
ResourceName_s The unique identifier for a machine within the workspace
ComputerName_s The computer FQDN
Ipv4Addresses_s A list of the server's IPv4 addresses
Ipv6Addresses_s A list of the server's IPv6 addresses
DnsNames_s An array of DNS names
OperatingSystemFamily_s Windows or Linux
OperatingSystemFullName_s The full name of the operating system
Bitness_s The bitness of the machine (32-bit or 64-bit)
PhysicalMemory_d The physical memory in MB
Cpus_d The number of CPUs
CpuSpeed_d The CPU speed in MHz
VirtualizationState_s unknown, physical, virtual, hypervisor
VirtualMachineType_s hyperv, vmware, and so on
VirtualMachineNativeMachineId_g The VM ID as assigned by its hypervisor
VirtualMachineName_s The name of the VM
BootTime_t The boot time

ServiceMapProcess_CL Type records

Records with a type of ServiceMapProcess_CL have inventory data for TCP-connected processes on servers with the Dependency agent. These records have the properties in the following table:

Property Description
Type ServiceMapProcess_CL
SourceSystem OpsManager
ResourceId The unique identifier for a process within the workspace
ResourceName_s The unique identifier for a process within the machine on which it is running
MachineResourceName_s The resource name of the machine
ExecutableName_s The name of the process executable
StartTime_t The process pool start time
FirstPid_d The first PID in the process pool
Description_s The process description
CompanyName_s The name of the company
InternalName_s The internal name
ProductName_s The name of the product
ProductVersion_s The product version
FileVersion_s The file version
CommandLine_s The command line
ExecutablePath _s The path to the executable file
WorkingDirectory_s The working directory
UserName The account under which the process is executing
UserDomain The domain under which the process is executing

Sample log searches

List all known machines

ServiceMapComputer_CL | summarize arg_max(TimeGenerated, *) by ResourceId

List the physical memory capacity of all managed computers.

ServiceMapComputer_CL | summarize arg_max(TimeGenerated, *) by ResourceId | project PhysicalMemory_d, ComputerName_s

List computer name, DNS, IP, and OS.

ServiceMapComputer_CL | summarize arg_max(TimeGenerated, *) by ResourceId | project ComputerName_s, OperatingSystemFullName_s, DnsNames_s, Ipv4Addresses_s

Find all processes with "sql" in the command line

ServiceMapProcess_CL | where CommandLine_s contains_cs "sql" | summarize arg_max(TimeGenerated, *) by ResourceId

Find a machine (most recent record) by resource name

search in (ServiceMapComputer_CL) "m-4b9c93f9-bc37-46df-b43c-899ba829e07b" | summarize arg_max(TimeGenerated, *) by ResourceId

Find a machine (most recent record) by IP address

search in (ServiceMapComputer_CL) "10.229.243.232" | summarize arg_max(TimeGenerated, *) by ResourceId

List all known processes on a specified machine

ServiceMapProcess_CL | where MachineResourceName_s == "m-559dbcd8-3130-454d-8d1d-f624e57961bc" | summarize arg_max(TimeGenerated, *) by ResourceId

List all computers running SQL

ServiceMapComputer_CL | where ResourceName_s in ((search in (ServiceMapProcess_CL) "\*sql\*" | distinct MachineResourceName_s)) | distinct ComputerName_s

List all unique product versions of curl in my datacenter

ServiceMapProcess_CL | where ExecutableName_s == "curl" | distinct ProductVersion_s

Create a computer group of all computers running CentOS

ServiceMapComputer_CL | where OperatingSystemFullName_s contains_cs "CentOS" | distinct ComputerName_s

Summarize the outbound connections from a group of machines

// the machines of interest
let machines = datatable(m: string) ["m-82412a7a-6a32-45a9-a8d6-538354224a25"];
// map of ip to monitored machine in the environment
let ips=materialize(ServiceMapComputer_CL
| summarize ips=makeset(todynamic(Ipv4Addresses_s)) by MonitoredMachine=ResourceName_s
| mvexpand ips to typeof(string));
// all connections to/from the machines of interest
let out=materialize(VMConnection
| where Machine in (machines)
| summarize arg_max(TimeGenerated, *) by ConnectionId);
// connections to localhost augmented with RemoteMachine
let local=out
| where RemoteIp startswith "127."
| project ConnectionId, Direction, Machine, Process, ProcessName, SourceIp, DestinationIp, DestinationPort, Protocol, RemoteIp, RemoteMachine=Machine;
// connections not to localhost augmented with RemoteMachine
let remote=materialize(out
| where RemoteIp !startswith "127."
| join kind=leftouter (ips) on $left.RemoteIp == $right.ips
| summarize by ConnectionId, Direction, Machine, Process, ProcessName, SourceIp, DestinationIp, DestinationPort, Protocol, RemoteIp, RemoteMachine=MonitoredMachine);
// the remote machines to/from which we have connections
let remoteMachines = remote | summarize by RemoteMachine;
// all augmented connections
(local)
| union (remote)
//Take all outbound records but only inbound records that come from either //unmonitored machines or monitored machines not in the set for which we are computing dependencies.
| where Direction == 'outbound' or (Direction == 'inbound' and RemoteMachine !in (machines))
| summarize by ConnectionId, Direction, Machine, Process, ProcessName, SourceIp, DestinationIp, DestinationPort, Protocol, RemoteIp, RemoteMachine
// identify the remote port
| extend RemotePort=iff(Direction == 'outbound', DestinationPort, 0)
// construct the join key we'll use to find a matching port
| extend JoinKey=strcat_delim(':', RemoteMachine, RemoteIp, RemotePort, Protocol)
// find a matching port
| join kind=leftouter (VMBoundPort 
| where Machine in (remoteMachines) 
| summarize arg_max(TimeGenerated, *) by PortId 
| extend JoinKey=strcat_delim(':', Machine, Ip, Port, Protocol)) on JoinKey
// aggregate the remote information
| summarize Remote=makeset(iff(isempty(RemoteMachine), todynamic('{}'), pack('Machine', RemoteMachine, 'Process', Process1, 'ProcessName', ProcessName1))) by ConnectionId, Direction, Machine, Process, ProcessName, SourceIp, DestinationIp, DestinationPort, Protocol

Next steps