Chapter 6. Reference Information
The content in this section is derived from the engineering documentation for this image. It is provided for reference as it can be useful for development purposes and for testing beyond the scope of the product documentation.
6.1. Persistent Templates
The JBoss EAP database templates, which deploy JBoss EAP and database pods, have both ephemeral and persistent variations. For example, for a JBoss EAP application backed by a MongoDB database, there are eap70-mongodb-s2i and eap70-mongodb-persistent-s2i templates.
Persistent templates include an environment variable to provision a persistent volume claim, which binds with an available persistent volume to be used as a storage volume for the JBoss EAP for OpenShift deployment. Information, such as timer schema, log handling, or data updates, is stored on the storage volume, rather than in ephemeral container memory. This information persists if the pod goes down for any reason, such as project upgrade, deployment rollback, or an unexpected error.
Without a persistent storage volume for the deployment, this information is stored in the container memory only, and is lost if the pod goes down for any reason.
For example, an EE timer backed by persistent storage continues to run if the pod is restarted. Any events triggered by the timer during the restart process are enacted when the application is running again.
Conversely, if the EE timer is running in the container memory, the timer status is lost if the pod is restarted, and starts from the beginning when the pod is running again.
6.2. Information Environment Variables
The following environment variables are designed to provide information to the image and should not be modified by the user:
| Variable Name | Description and Value |
|---|---|
| JBOSS_IMAGE_NAME | The image name.
Value: |
| JBOSS_IMAGE_RELEASE | The image release label.
Value: |
| JBOSS_IMAGE_VERSION | The image version.
Value: |
| JBOSS_MODULES_SYSTEM_PKGS | A comma-separated list of JBoss EAP system modules packages that are available to applications.
Value: |
| STI_BUILDER |
Provides OpenShift S2I support for
Value: |
6.3. Configuration Environment Variables
You can configure the following environment variables to adjust the image without requiring a rebuild.
| Variable Name | Description |
|---|---|
| AB_JOLOKIA_AUTH_OPENSHIFT |
Switch on client authentication for OpenShift TLS communication. The value of this parameter can be
|
| AB_JOLOKIA_CONFIG | If set, uses this fully qualified file path for the Jolokia JVM agent properties, which are described in Jolokia’s reference manual.
If not set, the
Example value: |
| AB_JOLOKIA_DISCOVERY_ENABLED | Enable Jolokia discovery.
Defaults to |
| AB_JOLOKIA_HOST | Host address to bind to.
Defaults to
Example value: |
| AB_JOLOKIA_HTTPS | Switch on secure communication with HTTPS.
By default self-signed server certificates are generated if no
Example value: |
| AB_JOLOKIA_ID | Agent ID to use.
The default value is the
Example value: |
| AB_JOLOKIA_OFF |
If set to Jolokia is enabled by default. |
| AB_JOLOKIA_OPTS |
Additional options to be appended to the agent configuration. They should be given in the format
Example value: |
| AB_JOLOKIA_PASSWORD | The password for basic authentication. By default, authentication is switched off.
Example value: |
| AB_JOLOKIA_PASSWORD_RANDOM |
Determines if a random
Set to |
| AB_JOLOKIA_PORT | The port to listen to.
Defaults to
Example value: |
| AB_JOLOKIA_USER | The name of the user to use for basic authentication.
Defaults to
Example value: |
| CLI_GRACEFUL_SHUTDOWN |
If set to any non-zero length value, the image will prevent shutdown with the
Example value: |
| CONTAINER_HEAP_PERCENT | Set the maximum Java heap size, as a percentage of available container memory.
Example value: |
| CUSTOM_INSTALL_DIRECTORIES | A list of comma-separated directories used for installation and configuration of artifacts for the image during the S2I process.
Example value: |
| DEFAULT_JMS_CONNECTION_FACTORY |
This value is used to specify the default JNDI binding for the JMS connection factory, for example
Example value: |
| ENABLE_ACCESS_LOG | Enable logging of access messages to the standard output channel. Logging of access messages is implemented using following methods:
Defaults to |
| INITIAL_HEAP_PERCENT | Set the initial Java heap size, as a percentage of the maximum heap size.
Example value: |
| JAVA_OPTS_APPEND | Server startup options.
Example value: |
| JBOSS_MODULES_SYSTEM_PKGS_APPEND |
A comma-separated list of package names that will be appended to the
Example value: |
| JGROUPS_PING_PROTOCOL |
JGroups protocol to use for node discovery. Can be either |
| MQ_SIMPLE_DEFAULT_PHYSICAL_DESTINATION |
For backwards compatibility, set to |
| OPENSHIFT_DNS_PING_SERVICE_NAME | Name of the service exposing the ping port on the servers for the DNS discovery mechanism.
Example value: |
| OPENSHIFT_DNS_PING_SERVICE_PORT |
The port number of the ping port for the DNS discovery mechanism. If not specified, an attempt will be made to discover the port number from the SRV records for the service, otherwise the default
Defaults to |
| OPENSHIFT_KUBE_PING_LABELS | Clustering labels selector for the Kubernetes discovery mechanism.
Example value: |
| OPENSHIFT_KUBE_PING_NAMESPACE | Clustering project namespace for the Kubernetes discovery mechanism.
Example value: |
| SCRIPT_DEBUG |
If set to |
Other environment variables not listed above that can influence the product can be found in the JBoss EAP documentation.
6.4. Application Templates
| Variable Name | Description |
|---|---|
| AUTO_DEPLOY_EXPLODED | Controls whether exploded deployment content should be automatically deployed.
Example value: |
6.5. Exposed Ports
| Port Number | Description |
|---|---|
| 8443 | HTTPS |
| 8778 | Jolokia Monitoring |
6.6. Datasources
Datasources are automatically created based on the value of some of the environment variables.
The most important environment variable is DB_SERVICE_PREFIX_MAPPING, as it defines JNDI mappings for the datasources. The allowed value for this variable is a comma-separated list of POOLNAME-DATABASETYPE=PREFIX triplets, where:
-
POOLNAMEis used as thepool-namein the datasource. -
DATABASETYPEis the database driver to use. -
PREFIXis the prefix used in the names of environment variables that are used to configure the datasource.
6.6.1. JNDI Mappings for Datasources
For each POOLNAME-DATABASETYPE=PREFIX triplet defined in the DB_SERVICE_PREFIX_MAPPING environment variable, the launch script creates a separate datasource, which is executed when running the image.
The first part (before the equal sign) of the DB_SERVICE_PREFIX_MAPPING should be lowercase.
The DATABASETYPE determines the driver for the datasource. Currently, only postgresql and mysql are supported.
Do not use any special characters for the POOLNAME parameter.
6.6.1.1. Database Drivers
Every image contains Java drivers for MySQL, PostgreSQL and MongoDB databases deployed. Datasources are generated only for MySQL and PostgreSQL databases.
For MongoDB database there are no JNDI mappings created because MongoDB is not a SQL database.
6.6.1.2. Datasource Configuration Environment Variables
To configure other datasource properties, use the following environment variables.
Be sure to replace the values for POOLNAME, DATABASETYPE, and PREFIX in the following variable names with the appropriate values. These replaceable values are described in this section and in the Datasources section.
| Variable Name | Description |
|---|---|
| POOLNAME_DATABASETYPE_SERVICE_HOST |
Defines the database server’s host name or IP address to be used in the datasource’s
Example value: |
| POOLNAME_DATABASETYPE_SERVICE_PORT | Defines the database server’s port for the datasource.
Example value: |
| PREFIX_BACKGROUND_VALIDATION |
When set to |
| PREFIX_BACKGROUND_VALIDATION_MILLIS |
Specifies frequency of the validation, in milliseconds, when the |
| PREFIX_CONNECTION_CHECKER | Specifies a connection checker class that is used to validate connections for the particular database in use.
Example value: |
| PREFIX_DATABASE | Defines the database name for the datasource.
Example value: |
| PREFIX_DRIVER | Defines Java database driver for the datasource.
Example value: |
| PREFIX_EXCEPTION_SORTER | Specifies the exception sorter class that is used to properly detect and clean up after fatal database connection exceptions.
Example value: |
| PREFIX_JNDI |
Defines the JNDI name for the datasource. Defaults to
Example value: |
| PREFIX_JTA | Defines Java Transaction API (JTA) option for the non-XA datasource. The XA datasources are already JTA capable by default.
Defaults to |
| PREFIX_MAX_POOL_SIZE | Defines the maximum pool size option for the datasource.
Example value: |
| PREFIX_MIN_POOL_SIZE | Defines the minimum pool size option for the datasource.
Example value: |
| PREFIX_NONXA |
Defines the datasource as a non-XA datasource. Defaults to |
| PREFIX_PASSWORD | Defines the password for the datasource.
Example value: |
| PREFIX_TX_ISOLATION | Defines the java.sql.Connection transaction isolation level for the datasource.
Example value: |
| PREFIX_URL | Defines connection URL for the datasource.
Example value: |
| PREFIX_USERNAME | Defines the username for the datasource.
Example value: |
When running this image in OpenShift, the POOLNAME_DATABASETYPE_SERVICE_HOST and POOLNAME_DATABASETYPE_SERVICE_PORT environment variables are set up automatically from the database service definition in the OpenShift application template, while the others are configured in the template directly as env entries in container definitions under each pod template.
6.6.1.3. Examples
These examples show how value of the DB_SERVICE_PREFIX_MAPPING environment variable influences datasource creation.
6.6.1.3.1. Single Mapping
Consider value test-postgresql=TEST.
This creates a datasource with java:jboss/datasources/test_postgresql name. Additionally, all the required settings like password and username are expected to be provided as environment variables with the TEST_ prefix, for example TEST_USERNAME and TEST_PASSWORD.
6.6.1.3.2. Multiple Mappings
You can specify multiple database mappings.
Always separate multiple datasource mappings with a comma.
Consider the following value for the DB_SERVICE_PREFIX_MAPPING environment variable: cloud-postgresql=CLOUD,test-mysql=TEST_MYSQL.
This creates the following two datasources:
-
java:jboss/datasources/test_mysql -
java:jboss/datasources/cloud_postgresql
Then you can use TEST_MYSQL prefix for configuring things like the username and password for the MySQL datasource, for example TEST_MYSQL_USERNAME. And for the PostgreSQL datasource, use the CLOUD_ prefix, for example CLOUD_USERNAME.
6.7. Clustering
JBoss EAP clustering on OpenShift is achieved through one of two discovery mechanisms: Kubernetes or DNS.
This is done by configuring the JGroups protocol stack in the standalone-openshift.xml configuration file with either the <openshift.KUBE_PING/> or <openshift.DNS_PING/> element. To use an environment variable to specify the discovery mechanism for the JBoss EAP for OpenShift image, set JGROUPS_PING_PROTOCOL on the image deployment to either openshift.KUBE_PING or openshift.DNS_PING.
The openshift.KUBE_PING discovery mechanism is the default mechanism when provisioning an application on top of the JBoss EAP for OpenShift image directly. However, the openshift.DNS_PING is the default discovery mechanism when using one of the available application templates to deploy an application on top of the JBoss EAP for OpenShift image.
The openshift.DNS_PING and openshift.KUBE_PING discovery mechanisms are not compatible with each other. It is not possible to form a supercluster out of two independent child clusters, with one using the openshift.DNS_PING mechanism for discovery and the other using the openshift.KUBE_PING mechanism. Similarly, when performing a rolling upgrade, the discovery mechanism needs to be identical for both the source and the target clusters.
6.7.1. Configuring KUBE_PING
For KUBE_PING to work, the following steps must be taken:
The JGroups protocol stack must be configured to use
KUBE_PINGas the discovery mechanism.You can do this by setting the
JGROUPS_PING_PROTOCOLenvironment variable toopenshift.KUBE_PING:JGROUPS_PING_PROTOCOL=openshift.KUBE_PING
The
OPENSHIFT_KUBE_PING_NAMESPACEenvironment variable must be set to your OpenShift project name. If not set, the server behaves as a single-node cluster (a "cluster of one"). For example:OPENSHIFT_KUBE_PING_NAMESPACE=PROJECT_NAMEThe
OPENSHIFT_KUBE_PING_LABELSenvironment variable should be set. This should match the label set at the service level. If not set, pods outside of your application (albeit in your namespace) will try to join. For example:OPENSHIFT_KUBE_PING_LABELS=app=APP_NAMEAuthorization must be granted to the service account the pod is running under to be allowed to access Kubernetes' REST API. This is done using the OpenShift CLI. The following example uses the
defaultservice account in the current project’s namespace:oc policy add-role-to-user view system:serviceaccount:$(oc project -q):default -n $(oc project -q)
Using the
eap-service-accountin the project namespace:oc policy add-role-to-user view system:serviceaccount:$(oc project -q):eap-service-account -n $(oc project -q)
See Prepare OpenShift for Application Deployment for more information on adding policies to service accounts.
6.7.2. Configuring DNS_PING
For DNS_PING to work, the following steps must be taken:
The JGroups protocol stack must be configured to use
DNS_PINGas the discovery mechanism.You can do this by setting the
JGROUPS_PING_PROTOCOLenvironment variable toopenshift.DNS_PING:JGROUPS_PING_PROTOCOL=openshift.DNS_PING
The
OPENSHIFT_DNS_PING_SERVICE_NAMEenvironment variable must be set to the name of the ping service for the cluster. If not set, the server will act as if it is a single-node cluster (a "cluster of one").OPENSHIFT_DNS_PING_SERVICE_NAME=PING_SERVICE_NAMEThe
OPENSHIFT_DNS_PING_SERVICE_PORTenvironment variable should be set to the port number on which the ping service is exposed. TheDNS_PINGprotocol attempts to discern the port from the SRV records, otherwise it defaults to8888.OPENSHIFT_DNS_PING_SERVICE_PORT=PING_PORTA ping service which exposes the ping port must be defined. This service should be headless (ClusterIP=None) and must have the following:
- The port must be named.
The service must be annotated with
service.alpha.kubernetes.io/tolerate-unready-endpointsset to"true".NoteOmitting this annotation will result in each node forming their own "cluster of one" during startup, then merging their cluster into the other nodes' clusters after startup, as the other nodes are not detected until after they have started.
kind: Service apiVersion: v1 spec: clusterIP: None ports: - name: ping port: 8888 selector: deploymentConfig: eap-app metadata: name: eap-app-ping annotations: service.alpha.kubernetes.io/tolerate-unready-endpoints: "true" description: "The JGroups ping port for clustering."
DNS_PING does not require any modifications to the service account and works using the default permissions.
6.8. Security Domains
To configure a new Security Domain, the user must define the SECDOMAIN_NAME environment variable.
This results in the creation of a security domain named after the environment variable. The user may also define the following environment variables to customize the domain:
| Variable name | Description |
|---|---|
| SECDOMAIN_NAME | Defines an additional security domain.
Example value: |
| SECDOMAIN_PASSWORD_STACKING |
If defined, the
Example value: |
| SECDOMAIN_LOGIN_MODULE | The login module to be used.
Defaults to |
| SECDOMAIN_USERS_PROPERTIES | The name of the properties file containing user definitions.
Defaults to |
| SECDOMAIN_ROLES_PROPERTIES | The name of the properties file containing role definitions.
Defaults to |
6.9. HTTPS Environment Variables
| Variable name | Description |
|---|---|
| HTTPS_NAME |
If defined along with
Example value: |
| HTTPS_PASSWORD |
If defined along with
Example value: |
| HTTPS_KEYSTORE |
If defined along with
Example value: |
6.10. Administration Environment Variables
| Variable name | Description |
|---|---|
| ADMIN_USERNAME |
If both this and
Example value: |
| ADMIN_PASSWORD |
The password for the specified
Example value: |
6.11. S2I
The image includes S2I scripts and Maven.
Maven is currently only supported as a build tool for applications that are supposed to be deployed on JBoss EAP-based containers (or related/descendant images) on OpenShift.
Only WAR deployments are supported at this time.
6.11.1. Custom Configuration
It is possible to add custom configuration files for the image. All files put into configuration/ directory will be copied into EAP_HOME/standalone/configuration/. For example to override the default configuration used in the image, just add a custom standalone-openshift.xml into the configuration/ directory. See example for such a deployment.
6.11.1.1. Custom Modules
It is possible to add custom modules. All files from the modules/ directory will be copied into EAP_HOME/modules/. See example for such a deployment.
6.11.2. Deployment Artifacts
By default, artifacts from the source target directory will be deployed. To deploy from different directories set the ARTIFACT_DIR environment variable in the BuildConfig definition. ARTIFACT_DIR is a comma-delimited list. For example: ARTIFACT_DIR=app1/target,app2/target,app3/target
6.11.3. Artifact Repository Mirrors
A repository in Maven holds build artifacts and dependencies of various types, for example, all of the project JARs, library JARs, plug-ins, or any other project specific artifacts. It also specifies locations from where to download artifacts while performing the S2I build. Besides using central repositories, it is a common practice for organizations to deploy a local custom mirror repository.
Benefits of using a mirror are:
- Availability of a synchronized mirror, which is geographically closer and faster.
- Ability to have greater control over the repository content.
- Possibility to share artifacts across different teams (developers, CI), without the need to rely on public servers and repositories.
- Improved build times.
Often, a repository manager can serve as local cache to a mirror. Assuming that the repository manager is already deployed and reachable externally at http://10.0.0.1:8080/repository/internal/, the S2I build can then use this manager by supplying the MAVEN_MIRROR_URL environment variable to the build configuration of the application as follows:
Identify the name of the build configuration to apply
MAVEN_MIRROR_URLvariable against.oc get bc -o name buildconfig/eap
Update build configuration of
eapwith aMAVEN_MIRROR_URLenvironment variable.oc env bc/eap MAVEN_MIRROR_URL="http://10.0.0.1:8080/repository/internal/" buildconfig "eap" updated
Verify the setting.
oc env bc/eap --list # buildconfigs eap MAVEN_MIRROR_URL=http://10.0.0.1:8080/repository/internal/
- Schedule new build of the application.
During application build, you will notice that Maven dependencies are pulled from the repository manager, instead of the default public repositories. Also, after the build is finished, you will see that the mirror is filled with all the dependencies that were retrieved and used during the build.
6.11.4. Scripts
run-
This script uses the
openshift-launch.shscript that configures and starts JBoss EAP with thestandalone-openshift.xmlconfiguration. assemble-
This script uses Maven to build the source, create a package (WAR), and move it to the
EAP_HOME/standalone/deploymentsdirectory.
6.11.5. Environment Variables
You can influence the way the build is executed by supplying environment variables to the s2i build command. The environment variables that can be supplied are:
| Variable name | Description |
|---|---|
| ARTIFACT_DIR |
The
Example value: |
| HTTP_PROXY_HOST | Host name or IP address of a HTTP proxy for Maven to use.
Example value: |
| HTTP_PROXY_PORT | TCP Port of a HTTP proxy for Maven to use.
Example value: |
| HTTP_PROXY_USERNAME |
If supplied with
Example value: |
| HTTP_PROXY_PASSWORD |
If supplied with
Example value: |
| HTTP_PROXY_NONPROXYHOSTS | If supplied, a configured HTTP proxy will ignore these hosts.
Example value: |
| MAVEN_ARGS | Overrides the arguments supplied to Maven during build.
Example value: |
| MAVEN_ARGS_APPEND | Appends user arguments supplied to Maven during build.
Example value: |
| MAVEN_MIRROR_URL | URL of a Maven Mirror/repository manager to configure.
Example value: |
| MAVEN_CLEAR_REPO | Optionally clear the local Maven repository after the build.
Example value: |
| APP_DATADIR | If defined, directory in the source from where data files are copied.
Example value: |
| DATA_DIR |
Directory in the image where data from
Example value: |
For more information, see Build and Run a Java Application on the JBoss EAP for OpenShift Image, which uses Maven and the S2I scripts included in the JBoss EAP for OpenShift image.
6.12. SSO
This image contains support for Red Hat JBoss SSO-enabled applications.
See the Red Hat JBoss SSO for OpenShift documentation for more information on how to deploy the Red Hat JBoss SSO for OpenShift image with the JBoss EAP for OpenShift image.
| Variable name | Description |
|---|---|
| SSO_URL | URL of the SSO server. |
| SSO_REALM | SSO realm for the deployed applications. |
| SSO_PUBLIC_KEY | Public key of the SSO Realm. This field is optional but if omitted can leave the applications vulnerable to man-in-middle attacks. |
| SSO_USERNAME | SSO User required to access the SSO REST API.
Example value: |
| SSO_PASSWORD |
Password for the SSO user defined by the
Example value: |
| SSO_SAML_KEYSTORE |
Keystore location for SAML. Defaults to |
| SSO_SAML_KEYSTORE_PASSWORD |
Keystore password for SAML. Defaults to |
| SSO_SAML_CERTIFICATE_NAME |
Alias for keys/certificate to use for SAML. Defaults to |
| SSO_BEARER_ONLY | SSO Client Access Type. (Optional)
Example value: |
| SSO_CLIENT |
Path for SSO redirects back to the application. Defaults to match |
| SSO_ENABLE_CORS |
If |
| SSO_SECRET | The SSO Client Secret for Confidential Access.
Example value: |
| SSO_SECURE_SSL_CONNECTIONS |
If |
6.13. Transaction Recovery
When a cluster is scaled down, it is possible for transaction branches to be in doubt. There is a technology preview automated recovery pod that is meant to complete these branches, but there are rare scenarios, such as a network split, where the recovery may fail. In these cases, manual transaction recovery might be necessary.
6.13.1. Unsupported Transaction Recovery Scenarios
JTS transactions
Because the network endpoint of the parent is encoded in recovery coordinator IORs, recovery cannot work reliably if either the child or parent node recovers with either a new IP address, or if it is intended to be accessed using a virtualized IP address.
XTS transactions
XTS does not work in a clustered scenario for recovery purposes. See JBTM-2742 for details.
- Transactions propagated over JBoss Remoting
Transactions propagated over XATerminator
Because the EIS is intended to be connected to a single instance of a Java EE application server, there are no well-defined ways to couple these processes.
6.13.2. Manual Transaction Recovery Process
The goal of the following procedure is to find and manually resolve in-doubt branches in cases where automated recovery has failed.
6.13.2.1. Caveats
This procedure only describes how to manually recover transactions that were wholly self-contained within a single JVM. The procedure does not describe how to recover JTA transactions that have been propagated to other JVMs.
There are various network partition scenarios in which OpenShift might start multiple instances of the same pod with the same IP address and same node name and where, due to the partition, the old pod is still running. During manual recovery, this might result in a situation where you might be connected to a pod that has a stale view of the object store. If you think you are in this scenario, it is recommended that all JBoss EAP pods be shut down to ensure that none of the resource managers or object stores are in use.
When you enlist a resource in an XA transaction, it is your responsibility to ensure that each resource type is supported for recovery. For example, it is known that PostgreSQL and MySQL are well-behaved with respect to recovery, but for others, such as A-MQ and JDV resource managers, you should check documentation of the specific OpenShift release.
The deployment must use a JDBC object store.
The transaction manager relies on the uniqueness of node identifiers. The maximum byte length of an XID is set by the XA specification and cannot be changed. Due to the data that the JBoss EAP for OpenShift image must include in the XID, this leaves room for 23 bytes in the node identifier.
OpenShift coerces the node identifier to fit this 23 byte limit:
-
For all node names, even those under 23 bytes, the
-(dash) character is stripped out. - If the name is still over 23 bytes, characters are truncated from the beginning of the name until length of the name is within the 23 byte limit.
However, this process might impact the uniqueness of the identifier. For example, the names aaa123456789012345678m0jwh and bbb123456789012345678m0jwh are both truncated to 123456789012345678m0jwh, which breaks the uniqueness of the names that are expected. In another example, this-pod-is-m0jwh and thispod-is-m0jwh are both truncated to thispodism0jwh, again breaking the uniqueness of the names.
It is your responsibility to ensure that the node names you configure are unique, keeping in mind the above truncation process.
6.13.2.2. Prerequisite
It is assumed the OpenShift instance has been configured with a JDBC store, and that the store tables are partitioned using a table prefix corresponding to the pod name. This should be automatic whenever a JBoss EAP deployment is in use. This is different from the automated recovery example, which uses a file store with split directories on a shared volume. You can verify that the JBoss EAP instance is using a JDBC object store by looking at the configuration of the transactions subsystem in a running pod:
Determine if the
/opt/eap/standalone/configuration/openshift-standalone.xmlconfiguration file contains an element for the transaction subsystem:<subsystem xmlns="urn:jboss:domain:transactions:3.0">
If the JDBC object store is in use, then there is an entry similar to the following:
<jdbc-store datasource-jndi-name="java:jboss/datasources/jdbcstore_postgresql"/>
NoteThe JNDI name identifies the datasource used to store the transaction logs.
6.13.2.3. Procedure
The following procedure details the process of manual transaction recovery solely for datasources.
- Use the database vendor tooling to list the XIDs (transaction branch identifiers) for in-doubt branches. It is necessary to list XIDs for all datasources that were in use by any deployments running on the pod that failed or was scaled down. Refer to the vendor documentation for the database product in use.
For each such XID, determine which pod created the transaction and check to see if that pod is still running.
- If it is running, then leave the branch alone.
If the pod is not running, assume it was removed from the cluster and you must apply the manual resolution procedure described here. Look in the transaction log storage that was used by the failed pod to see if there is a corresponding transaction log:
- If there is a log, then manually commit the XID using the vendor tooling.
- If there is not a log, assume it is an orphaned branch and roll back the XID using the vendor tooling.
The rest of this procedure explains in detail how to carry out each of these steps.
6.13.2.3.1. Resolving In-doubt Branches
First, find all the resources that the deployment is using.
It is recommended that you do this using the JBoss EAP managagement CLI. Although the resources should be defined in the JBoss EAP standalone-openshift.xml configuration file, there are other ways they can be made available to the transaction subsystem within the application server. For example, this can be done using a file in a deployment, or dynamically using the management CLI at runtime.
- Open a terminal on a pod running a JBoss EAP instance in the cluster of the failed pod. If there is no such pod, scale up to one.
-
Create a management user using the
/opt/eap/bin/add-user.shscript. -
Log into the management CLI using the
/opt/eap/bin/jboss-cli.shscript. List the datasources configured on the server. These are the ones that may contain in-doubt transaction branches.
/subsystem=datasources:read-resource { "outcome" => "success", "result" => { "data-source" => { "ExampleDS" => undefined, ... }, ... }Once you have the list, find the connection URL for each of the datasources. For example:
/subsystem=datasources/data-source=ExampleDS:read-attribute(name=connection-url) { "outcome" => "success", "result" => "jdbc:h2:mem:test;DB_CLOSE_DELAY=-1;DB_CLOSE_ON_EXIT=FALSE", "response-headers" => {"process-state" => "restart-required"} }Connect to each datasource and list any in-doubt transaction branches.
NoteThe table name that stores in-doubt branches will be different for each datasource vendor.
JBoss EAP has a default SQL query tool (H2) that you can use to check each database. For example:
java -cp /opt/eap/modules/system/layers/base/com/h2database/h2/main/h2-1.3.173.jar \ -url "jdbc:postgresql://localhost:5432/postgres" \ -user sa \ -password sa \ -sql "select gid from pg_prepared_xacts;"
Alternatively, you can use the resource’s native tooling. For example, for a PostGreSQL datasource called
sampledb, you can use the OpenShift client tools to remotely log in to the pod and query the in-doubt transaction table:$ oc rsh postgresql-2-vwf9n # rsh to the named pod sh-4.2$ psql sampledb psql (9.5.7) Type "help" for help. sampledb=# select gid from pg_prepared_xacts; 131077_AAAAAAAAAAAAAP//rBEAB440GK1aJ72oAAAAGHAtanRhLWNyYXNoLXJlYy0zLXAyY2N3_AAAAAAAAAAAAAP//rBEAB440GK1aJ72oAAAAGgAAAAEAAAAA
6.13.2.3.2. Extract the Global Transaction ID and Node Identifier from Each XID
When all XIDs for in-doubt branches are identified, convert the XIDs into a format that you can compare to the logs stored in the transaction tables of the transaction manager.
For example, the following Bash script can be used to perform this conversion. Assuming that $PG_XID holds the XID from the select statement above, then the JBoss EAP transaction ID can be obtained as follows:
PG_XID="$1"
IFS='_' read -ra lines <<< "$PG_XID"
[[ "${lines[0]}" = 131077 ]] || exit 0; # this script only works for our own FORMAT ID
PG_TID=${lines[1]}
a=($(echo "$PG_TID"| base64 -d | xxd -ps |tr -d '\n' | while read -N16 i ; do echo 0x$i ; done))
b=($(echo "$PG_TID"| base64 -d | xxd -ps |tr -d '\n' | while read -N8 i ; do echo 0x$i ; done))
c=("${b[@]:4}") # put the last 3 32-bit hexadecimal numbers into array c
# the negative elements of c need special handling since printf below only works with positive
# hexadecimal numbers
for i in "${!c[@]}"; do
arg=${c[$i]}
# inspect the MSB to see if arg is negative - if so convert it from a 2’s complement number
[[ $(($arg>>31)) = 1 ]] && x=$(echo "obase=16; $(($arg - 0x100000000 ))" | bc) || x=$arg
if [[ ${x:0:1} = \- ]] ; then # see if the first character is a minus sign
neg[$i]="-";
c[$i]=0x${x:1} # strip the minus sign and make it hex for use with printf below
else
neg[$i]=""
c[$i]=$x
fi
done
EAP_TID=$(printf %x:%x:${neg[0]}%x:${neg[1]}%x:${neg[2]}%x ${a[0]} ${a[1]} ${c[0]} ${c[1]} ${c[2]})
After completion, the $EAP_TID variable holds the global transaction ID of the transaction that created this XID. The node identifier of the pod that started the transaction is given by the output of the following bash command:
echo "$PG_TID"| base64 -d | tail -c +29
The node identifier starts from the 29th character of the PostgreSQL global transaction ID field.
- If this pod is still running, then leave this in-doubt branch alone since the transaction is still in flight.
If this pod is not running, then you need to search the relevant transaction log storage for the transaction log. The log storage is located in a JDBC table, which is named following the
os<node-identifier>jbosststxtablepattern.- If there is no such table, leave the branch alone as it is owned by some other transaction manager. The URL for the datasource containing this table is defined in the transaction subsystem description shown below.
If there is such a table, look for an entry that matches the global transaction ID.
- If there is an entry in the table that matches the global transaction ID, then the in-doubt branch needs to be committed using the datasource vendor tooling as described below.
- If there is no such entry, then the branch is an orphan and can safely be rolled back.
An example of how to commit an in-doubt PostgreSQL branch is shown below:
$ oc rsh postgresql-2-vwf9n sh-4.2$ psql sampledb psql (9.5.7) Type "help" for help. psql sampledb commit prepared '131077_AAAAAAAAAAAAAP//rBEAB440GK1aJ72oAAAAGHAtanRh ---- LWNyYXNoLXJlYy0zLXAyY2N3_AAAAAAAAAAAAAP//rBEAB440GK1aJ72oAAAAGgAAAAEAAAAA';
Repeat this procedure for all datasources and in-doubt branches.
6.13.2.3.3. Obtain the List of Node Identifiers of All Running JBoss EAP Instances in Any Cluster that Can Contact the Resource Managers
Node identifiers are configured to be the same name as the pod name. You can obtain the pod names in use using the oc command. Use the following command to list the running pods:
$ oc get pods | grep Running eap-manual-tx-recovery-app-4-26p4r 1/1 Running 0 23m postgresql-2-vwf9n 1/1 Running 0 41m
For each running pod, look in the output of the pod’s log and obtain the node name. For example, for first pod shown in the above output, use the following command:
$ oc logs eap-manual-tx-recovery-app-4-26p4r | grep "jboss.node.name" | head -1 jboss.node.name = tx-recovery-app-4-26p4r
The aforementioned JBoss node name identifier will always be truncated to the maximum length of 23 characters in total by removing characters from the beginning and retaining the trailing characters until the maximum length of 23 characters is reached.
6.13.2.3.4. Find the Transaction Logs
-
The transaction logs reside in a JDBC-backed object store. The JNDI name of this store is defined in the
transactionsubsystem definition of the JBoss EAP configuration file. - Look in the configuration file to find the datasource definition corresponding to the above JNDI name.
- Use the JNDI name to derive the connection URL.
You can use the URL to connect to the database and issue a
selectquery on the relevant in-doubt transaction table.Alternatively, if you know which pod the database is running on, and you know the name of the database, it might be easier to open an OpenShift remote shell into the pod and use the database tooling directly.
For example, if the JDBC store is hosted by a PostgreSQL database called
sampledbrunning on podpostgresql-2-vwf9n, then you can find the transaction logs using the following commands:NoteThe ostxrecoveryapp426p4rjbosststxtable table name listed in the following command has been chosen since it follows the pattern for JDBC table names holding the log storage entries. In your environment the table name will have similar form:
-
Starting with
osprefix. - The part in the middle is derived from the JBoss node name above, possibly deleting the "-" (dash) character if present.
-
Finally the
jbosststxtablesuffix is appended to create the final name of the table.
$ oc rsh postgresql-2-vwf9n sh-4.2$ psql sampledb psql (9.5.7) Type "help" for help. sampledb=# select uidstring from ostxrecoveryapp426p4rjbosststxtable where TYPENAME='StateManager/BasicAction/TwoPhaseCoordinator/AtomicAction' ; uidstring ------------------------------------- 0:ffff0a81009d:33789827:5a68b2bf:40 (1 row)-
Starting with
6.13.2.3.5. Cleaning Up the Transaction Logs for Reconciled In-doubt Branches
Do not delete the log unless you are certain that there are no remaining in-doubt branches.
When all the branches for a given transaction are complete, and all potential resources managers have been checked, including A-MQ and JDV, it is safe to delete the transaction log.
Issue the following command, specify the transaction log to be removed using the appropriate uidstring:
DELETE FROM ostxrecoveryapp426p4rjbosststxtable where uidstring = UIDSTRINGIf you do not delete the log, then completed transactions which failed after prepare, but which have now been resolved, will never be removed from the transaction log storage. The consequence of this is that unnecessary storage is used and future manual reconciliation will be more difficult.
6.14. Included JBoss Modules
The table below lists included JBoss Modules in the JBoss EAP for OpenShift image.
| JBoss Module |
|---|
| org.jboss.as.clustering.common |
| org.jboss.as.clustering.jgroups |
| org.jboss.as.ee |
| org.jboss.logmanager.ext |
| org.jgroups |
| org.mongodb |
| org.openshift.ping |
| org.postgresql |
| com.mysql |
| net.oauth.core |
Revised on 2018-01-24 22:28:00 EST
