PE known issues
Sections
- Initial agent run after upgrade can fail with many environments
- Installing or upgrading agents using TLSv1 fails with older OpenSSL versions
- Windows agent installation fails given a space in user name
- Converting legacy compilers fails with an external certificate authority
- Converted compilers can slow PuppetDB in multi-region installations
These are the known issues in PE 2021.2.
Installation and upgrade known issues
These are the known issues for installation and upgrade in this release.
Initial agent run after upgrade can fail with many environments
In installations with many environments, where file sync can take several minutes, the orchestration service fails to reload during the first post-upgrade Puppet run. As a workaround, re-run the Puppet agent until the orchestration service loads properly. To prevent encountering this error, you can clean up unused environments before upgrading, and wait several minutes after the installer completes to run the agent.
Installing or upgrading agents using TLSv1 fails with older OpenSSL versions
Script-based agent installation or upgrade on nodes that use TLSv1 can fail if the
curl version installed on the node uses OpenSSL earlier than version 1.0. This issue
produces an SSL error during any curl connection to the primary server. As a
workaround, add --ciphers AES256-SHA
to ~/.curlrc
so that curl calls always use an appropriate
cipher.
Windows agent installation fails given a space in user name
The Windows agent install script fails if executed
with a user name that includes a space, like Max
Spacey
. You receive the error Something went wrong with the
installation along with exit code 1639. As a workaround, run the install
script on Windows agents with a username without
spaces, or use an alternative installation method, such as the .msi
package.
Converting legacy compilers fails with an external certificate authority
puppet infrastructure
run convert_legacy_compiler
command fails with an error during the
certificate-signing step.
Agent_cert_regen: ERROR: Failed to regenerate agent certificate on node <compiler-node.domain.com> Agent_cert_regen: bolt/run-failure:Plan aborted: run_task 'enterprise_tasks::sign' failed on 1 target Agent_cert_regen: puppetlabs.sign/sign-cert-failed Could not sign request for host with certname <compiler-node.domain.com> using caserver <master-host.domain.com>
- Log on to the CA server and manually sign certificates for the compiler.
- On the compiler, run Puppet:
puppet agent -t
- Unpin the compiler from PE Master group, either from
the console, or from the CLI using the command:
/opt/puppetlabs/bin/puppet resource pe_node_group "PE Master" unpinned="<COMPILER_FQDN>"
- On your primary server, in the
pe.conf
file, remove the entrypuppet_enterprise::profile::database::private_temp_puppetdb_host
- If you have an external PE-PostgreSQL node, run Puppet on that node:
puppet agent -t
- Run Puppet on your primary server:
puppet agent -t
- Run Puppet on all compilers:
puppet agent -t
Converted compilers can slow PuppetDB in multi-region installations
In configurations that rely on high-latency connections between your primary servers and compilers – for example, in multi-region installations – converted compilers running the PuppetDB service might experience significant slowdowns. If your primary server and compilers are distributed among multiple data centers connected by high-latency links or congested network segments, reach out to Support for guidance before converting legacy compilers.
Disaster recovery known issues
These are the known issues for disaster recovery in this release.
An unreachable replica runs the primary server out of disk space
If a provisioned replica becomes unreachable, the associated primary server can quickly run out of disk space, causing a complete interruption to PE services. In larger installations, an outage can occur in under an hour. The disk usage issue is caused by the PE-PostgreSQL service on the primary server retaining change logs that the replica hasn't acknowledged.
puppet infrastructure forget
<REPLICA NODE NAME>
.Replica promotion can fail in air-gapped installations
If your primary server includes AIX or Solaris
pe_repo
classes, replica promotion fails in
air-gapped environments because the staged AIX and
Solaris tarballs aren't copied to the replica. As
a workaround, copy the staging directory /opt/puppetlabs/server/data/staging
from the primary server to the
replica before promotion.
FIPS known issues
These are the known issues with FIPS-enabled PE in this release.
Puppet Server FIPS installations don’t support Ruby’s OpenSSL module
FIPS-enabled PE installations don't support extensions
or modules that use the standard Ruby Open SSL
library, such as hiera-eyaml or the splunk_hec
module. As a workaround, you can use a non-FIPS-enabled primary server with
FIPS-enabled agents, which limits the issue to situations where only the agent uses
the Ruby library.
Errors when using puppet code
and puppet db
commands on FIPS-compliant platforms
pe-client-tools
packages are run on
FIPS-compliant platforms, puppet code
and puppet db
commands fail with SSL handshake errors. To
use puppet db
commands on a FIPS-compliant
platforms, install the puppetdb_cli
Ruby gem with the following command:
/opt/puppetlabs/puppet/bin/gem install puppetdb_cli --bindir /opt/puppetlabs/bin/
Copied!
To
use puppet code
commands on a FIPS-compliant
platforms, use the Code Manager API. Alternatively, you
can use pe-client-tools
on a non-FIPS-compliant
platform to access a FIPS-enabled primary server. Configuration and maintenance known issues
These are the known issues for configuration and maintenance in this release.
Restarting or running Puppet on infrastructure nodes can trigger an illegal reflective access operation warning
When restarting PE services or performing agent runs on infrastructure nodes, you might see the warning Illegal reflective access operation ... All illegal access operations will be denied in a future release in the command-line output or logs. These warnings are internal to PE service components, have no impact on their functionality, and can be safely disregarded.
Orchestration services known issues
These are the known issues for the orchestration services in this release.
Running plans during code deployment can result in failures
If a plan is running during a code deployment, things like compiling apply block statements or downloading and running tasks that are part of a plan might fail. This is because plans run on a combination of PE services, like orchestrator and puppetserver, and the code each service is acting on might get temporarily out of sync during a code deployment.
Console and console services known issues
These are the known issues for the console and console services in this release.
Gateway timeout errors in the console
Using facts to filter nodes might produce either a "502 Bad Gateway" or "Gateway Timeout" error instead of the expected results.
Patching known issues
These are the known issues for patching in this release.
Patching fails with excluded yum packages
In the patching task or plan, using yum_params
to
pass the --exclude
flag in order to exclude certain
packages can result in task or plan failure if the only packages requiring updates
are excluded. As a workaround, use the versionlock
command (which requires installing the yum-plugin-versionlock package) to lock the
packages you want to exclude at their current version. Alternatively, you can fix a
package at a particular version by specifying the version with a package resource
for a manifest that applies to the nodes to be patched.
Failed or in-progress reboots are incorrectly reporting that they finished rebooting successfully
When rebooting a node using the pe_patch::group_patching
plan, the check to detect if a node has
rebooted always detects that it finished rebooting successfully, even if the reboot
failed or is still in progress, due to a parsing error in the output. This behavior
has been observed and tested on RHEL-based platform
versions 6 and 7, and SLES version 12, but it might exist on other platforms as
well.
Code management known issues
These are the known issues for Code Manager, r10k, and file sync in this release.
Changing a file type in a control repo produces a checkout conflict error
Changing a file type in a control repository – for example, deleting a file and replacing it with a directory of the same name – generates the error JGitInternalException: Checkout conflict with files accompanied by a stack trace in the Puppet Server log. As a workaround, deploy the control repo with the original file deleted, and then deploy again with the replacement file or directory.
Enabling Code Manager and multithreading in Puppet Server deadlocks JRuby
Setting the new environment_timeout
parameter to any non-zero value – including the unlimited
default when Code Manager is enabled – interferes
with multithreading in Puppet Server and
can result in JRuby deadlocking after
several hours.
Code Manager and r10k do not identify the default branch for module repositories
When you use Code Manager or r10k to deploy modules from a Git
source, the default branch of the source repository is always
assumed to be main. If the module repository uses a default branch
that is not main, an error occurs. To work around this issue,
specify the default branch with the ref:
key in
your Puppetfile.