Agent running as Windows Service Limitations
When a TeamCity build agent is installed as a Windows service, there may appear various "Permission denied" or "Access denied" errors during the build process, see details below.
The user account used by the service is required to have sufficient permissions to perform the build and manage the service. If you run the TeamCity agent service under the SYSTEM account, do the following:
Change SYSTEM for a usual user account with necessary permissions granted.
Restart the service.
Windows service limitations
As a Windows service, the TeamCity agent and the build processes are not able to access network shares and mapped drives.
To overcome these restrictions, run TeamCity agent via console.
Issues with automated GUI and browser testing
These problems include errors running tests headless, issues with the interaction of the TeamCity agent with the Windows desktop, etc.
To resolve/ avoid these:
Run TeamCity agent via console.
Configure the build agent machine not to launch a screensaver locking the desktop.
Configure the TeamCity agent to start automatically (e.g. configure an automatic user logon on Windows start and then configure the TeamCity agent start (via agent.bat start) on the user logon). For graphical tests the build agent cannot be started as a service and it is recommended to configure the build agent launch with a 1 minute delay after the user auto-logon, e.g. using the "
bin\agent.bat start"command in the task scheduler and configuring the delay there.
Running automated GUI tests and using RDP
RDP uses its own video driver overriding the one from the machine's video card for the session. Redirecting the session to console will unload the Windows graphical drivers. This can be done by adding the following step to your build configuration prior to your tests (the example below is for PowerShell, but other languages (DOS, Python) can be used too):
quser [current username]" lists all the connections to that machine for the user, either console or graphical. The one listed as
rdp-tcp#* is the remote desktop connection which can be redirected to the console using "
tscon [connection id] /dest:console".
Issues with . Net Selenium
When a TeamCity agent is started as a Windows service and automated tests for .Net applications use Selenium WebDriver, the tests may fail due to browser drivers limitations. As a solution, consider starting the agent manually.
Early start of the service before other resources are initialized
To handle this, consider using the Automatic (Delayed Start) option of the service settings or configure service dependencies.
java.lang.OutOfMemoryError: unable to create new native thread error
If your TeamCity server is running on SUSE® Linux Enterprise (or using systemd Daemon), the java.lang.OutOfMemoryError: unable to create new native thread error may be caused by the cgroup process number controller feature limiting the number of processes and the amount of threads in a cgroup to 512 by default.
Increasing the limit (e.g. to 4096) on the TeamCity server should solve the issue.
See also this external posting.
Clearing Browser Caсhes
There is a web UI-related issue which some our users have encountered (and it cannot be reproduced on other computers) which is tied to the cached versions of content. If you have come across such problem, make sure your browser does not use cached versions of content by clearing browser caches.
Logging with Log4j in Your Tests
If you use Log4j logging in your tests, in some cases you may miss the Log4j output from your logs. In such cases please do the following:
Use Log4j 1.2.12
For Log4j 1.2.13+, add the
"Follow=true"parameter for the console appender used in Log4j configuration:<appender name="CONSOLE" class="org.apache.log4j.ConsoleAppender"> <param name="Follow" value="true"/> </appender>
Agent Service Can Exit on User Logout under Windows x64
The used version of Java Service Wrapper does not fully support Windows 64 and this causes agent launcher process to be killed on user logout. The agent itself will be function until the next restart (server upgrade or agent properties change).
Failed Build Can be Reported as a Successful One With Maven 2.0.7
This is a known bug in this version of Maven. Consider using any later version. In case it's not possible you can patch mvn.bat yourself by replacing the fragment at line 148 of
with the following one:
Most common indicators of conflicting software are errors like "Access is denied", "Permission denied" or java.io.FileNotFoundException mentioning the file that is present and is writable by the user the agent/build runs under. Also, certain software running in background (like antiviruses) can significantly slow down build agent operations like sources checkout, artifact publishing or even build running.
Certain antivirus software like Kaspersky Internet Security can result in Java process crashes or other misbehavior like inability to access files. e.g. see the issue. ESET antivirus can also slow down Ant/IntelliJ IDEA project builds a great deal (slowing down TCP connections to localhost on agent).
If you run antivirus on the TeamCity server or agent machines and get disk access errors or experience degraded performance, please disable or better completely uninstall the antivirus software before investigating the issue and reporting the issue to JetBrains.
It is recommended to exclude entire TeamCity server home and TeamCity Data Directory from the background checks and perform periodical checks there in the well-known maintenance window so that those do not affect server performance much. On TeamCity agent, it is recommended to exclude TeamCity agent home from the background checks.
Please disable various indexing services. e.g. there might be problems with Windows Indexing Service. See issue for more details. Windows System Restore Feature might also need disabling.
Please also do not install software with background indexing like WinCVS, TortoiseCVS, TortoiseSVN and other Tortoise* products. This applies to server and also to agents if you use agent-side checkout.
Skype software is known to:
use port 80 on the system so you might not be able to use TeamCity server using default 80 port.
corrupt layout of pages displayed in Internet Explorer. Internet Explorer Skype plugin is to blame. (TW-13052).
svn: E175002: Received fatal alert: bad_record_mac
Please add system property -Dsvnkit.http.sslProtocols=SSLv3,TLS on the build server (see Configuring TeamCity Server Startup Properties). If you use checkout on agent, add this property on build agent as well.
Subversion-related JVM Crashes
If JVM crashes while executing SVN-related code (e.g. under org.tmatesoft.svn package), you can try to disable it by either:
-Dsvnkit.useJNA=falseJVM option to the crashing process (server or agent), or
Making NTLM support less prioritative by passing
Anyway, upgrading the JVM used to the latest available version is recommended.
NUnit 2.4.6 Performance
Due to an issue in NUnit 2.4.6, its performance may be slower than NUnit 2.4.1. For additional information, please refer to the corresponding issue in our issue tracker: TW-4709
Using StarTeam SDK 9.0 instead of StarTeam SDK 9.3 on the TeamCity server can significantly improve VCS performance when there is a slow connection between TeamCity and StarTeam servers.
Perforce 2009.2 Performance on Windows
If you run Perforce 2009.2 on Windows you may experience significant slow down. This is an issue with P4 server running on Windows. Please refer to corresponding section in Perforce documentation.
Wrong times for build scheduled triggering (Timezone issues)
Please make sure you use the latest update for Java 8 installation available for your platform (e.g. OpenJDK 8 by AdoptOpenJDK) .
Upgrading IntelliJ IDEA May Affect Active Pre-Tested Commits
Before you upgrade to IntelliJ IDEA X (or other IntelliJ X platform products) please make sure you do not have active pre-tested commits, otherwise they will not be able to be committed after upgrade. This is only relevant if you use directory-based IDEA project (project files are stored under
Other Java Applications Running on the Same Server
If other web applications are available via the same hostname, a session cookie conflict can occur. This usually is visible via random user logouts or losing session-level data. (e.g. TW-12654). To resolve this, you can use different host names when accessing the applications.
The Server Does Not Start Claiming the Database is in Use
Only a single TeamCity server can work with one database, which is checked on the TeamCity server start. "The Database is in Use" error on the start-up is reported in either of the following cases:
An attempt to start more than one TeamCity server connected to the same database
A second TeamCity instance detected
The internal HSQL database is being used by another application
The error is most probably caused by the fact that there is another running TeamCity installation which is connected to the same database. Сheck that the database properties are correct and there is no other TeamCity server using the same database.
In TeamCity 8.0 and earlier, if all the settings are correct, the error can occur when the TeamCity server or the database server has been shut down incorrectly. The resolution depends on the database type:
MySQL: restart the MySQL server and then start TeamCity again.
PostgreSQL, Oracle, MS SQL: kill the connections from the incorrectly shut down TeamCity, and then start TeamCity again.
Internal database (HSQL): remove the
buildserver.lckfile from the
\systemdirectory, and then start TeamCity again.
Slow download from TeamCity server
If you experience slow speed when downloading artifacts from TeamCity, try checking the speed on the server machine, downloading from localhost. If the speed is OK for the localhost, the issue can be in the network configuration or OS/hardware settings when combined with TeamCity(Tomcat) settings.
Please also make sure <TeamCity Home>\conf\server.xml file corresponds to the file included in TeamCity distribution (can be checked in .tar.gz distribution). If you have the following "Connector" node (ports numbers can be different):
change it to:
and restart TeamCity server.
If this does not help with the download speed, to investigate the case you might need to find an administrator with appropriate network-related issues investigation skills to look into the case.
Failure to publish artifacts to server behind IIS reverse proxy
This problem is only relevant to configurations that involve IIS reverse proxy between build server and agents. Sometimes a build agent can be found in infinite loop trying to publish build artifacts to server. Build log looks like this:
meanwhile teamcity-agent.log is filled with 404 responses from IIS:
The most common cause for this is maxAllowedContentLength setting (in IIS) either
is set to too small value
is left unconfigured and so defaults to 30000000 bytes (<30 Mb)
So any artifact larger than maxAllowedContentLength is discarded by IIS Check the settings value and try to rerun your build
Prerelease packages are not visible in the TeamCity NuGet feed
Problem. Prerelease packages are not visible in the TeamCity NuGet feed.
Cause. NuGet clients prior to version 3 fail to list prerelease packages if the package version violates the required format.
Packages indexing is slow in TeamCity NuGet feed
Problem. After TeamCity server host machine move or upgrade to the TeamCity 2017.1 build metadata can be reset.
Cause. The TeamCity NuGet feed relies on build metadata, and packages re-indexing can take a lot of time depending on the number of packages and the idle time of the TeamCity server.
Solution. To speed up build metadata re-indexing, specify following internal properties:
To check the metadata indexing progress, look for lines similar to the ones below in the teamcity-server.log file:
After re-indexing is complete, remove these internal properties.
SSL problems when connecting to HTTPS from TeamCity (handshake alert: unrecognized_name)
This problem may happen when changing JVM from 1.6 to 1.7 and connecting some incorrectly configured HTTPS servers. The problem and workaround for it are described in this issue: http://youtrack.jetbrains.com/issue/TW-30210
SSL problems connecting MS SQL Server and TeamCity
Since MS SQL Server always establishes an SSL connection between the jdbc client (the TeamCity application) and the server, connection problems may occur ( SQL exception: Connection reset).
Affected MS SQL versions: Any version prior to the ones listed below:
SQL Server 2012 Production version and later
SQL Server 2008 Service Pack 3 Cumulative Update 4
SQL Server 2008R2 Service Pack 1 Cumulative Update 6
SQL Server 2008R2 Service Pack 2
Cause: The problem is caused by the Java 1.6 update addressing this security vulnerability.
Solution: Microsoft SQL Server upgrade is recommended.
Workarounds: If Microsoft SQL Server upgrade is not possible for some reason, TeamCity can be set up to use older Microsoft SQL Server versions with the database connection still SSL- or TLS-encrypted.
Continue using a block cipher such as
3DES_EDE_CBC, but disable CBC protection via
-Djsse.enableCBCProtection=falseJava command-line option (that can be added to
TEAMCITY_SERVER_OPTSenvironment variable, as described here). The
jsse.enableCBCProtectionJava system property is also available in all OpenJDK 8 versions and IBM J9 8.0.0 SR1 and later. Secure connection between TeamCity and Microsoft SQL Server would be stable but still vulnerable to CVE-2011-3389 also known as BEAST.
Fall back to a stream cipher (which is not susceptible to BEAST) such as
RC4_128. This will render the connection vulnerable to CVE-2015-2808.
Please try running with antivirus software uninstalled before reporting the issue to JetBrains. e.g. see the issue.
Windows Docker Containers
Since Windows 10 version 1803 with KB4340917 it's possible to use port mapping from containers to localhost. For previous Windows versions it works for the non-localhost IP address associated with this machine and you can access a running application via the machine's hostname or determine the IP address via the
ipconfigcommand. Note that the
netstat -ancommand may not show that the port is open on any IP address, while in fact it can work. This is also a known problem of Docker on Windows.
On Windows 10, the memory allocated per container is 1GB by default. To increase this value, use the following memory options:docker run ... -m 2GB -e TEAMCITY_SERVER_MEM_OPTS="-Xmx2g -XX:ReservedCodeCacheSize=350m"
On Windows 10 containers work via Hyper-V and may experience problems with network and other subsystems. To diagnose these problems, execute the following PowerShell script:Invoke-WebRequest https://aka.ms/Debug-ContainerHost.ps1 -UseBasicParsing | Invoke-Expression
When starting a TeamCity server from a Windows Docker image, make sure to grant
Authenticated UsersFull control over the directories used as volumes. See the related issue.
When starting a Windows Docker container with the directory C:/BuildAgent/work mapped as a volume to the container host, Git for Windows fails with a following error: "
Invalid path '/ContainerMappedDirectories': No such file or directory". The workaround is not to add "C:/BuildAgent/work" as a volume.
Information about installed Docker server OS on Windows missing on Agent
On Windows 10, the Docker server depends on Hyper-V service and its start may take a significant amount of time. To resolve the issue, configure the TCBuildAgent Windows service to depend on the Docker for Windows Service, "com.docker.service" by default.
Linux Docker Containers under Windows
Since TeamCity 2017.2, the Docker Wrapper works on Windows when Windows-based containers are started.
If a Linux container is started on a Windows machine, TeamCity displays the error message "Starting Linux Docker containers under Windows is not supported. To avoid this problem, add the teamcity.agent.jvm.os.name does not contain Windowsagent requirement.
If you need to support a use case when the Docker wrapper runs Linux containers under Windows platform, please vote for /comment on TW-51820.
Problems with local time in Windows containers
When using Windows Docker containers, there is an issue with incorrect time in Windows containers when system time in container goes out of sync with the time on the host machine. It could cause problems in integrations where response time is significant (e.g. OAuth tokens).
To address it, upgrade your host machine to Windows Server 2019 / Windows 10 1809 and use TeamCity docker images compatible with Windows containers 1809.
"Access is denied" or "Access to the path is denied" problem on container start
When docker is starting Windows containers with process isolation, it's using the SERVICE user account which lacks the write access to directory with docker volumes. To resolve it, grant the "Full control" permission to the SERVICE user for the "%PROGRAMDATA%\docker\volumes" directory.