Errata

Figure 4-8. Transformation of the query plan to a distributed query plan might be slightly misleading as stage number should go from large to lower number, with stage 0 in coordinator.
- trinodb.slack.com/archives/C01TEP0HJTH/p1675640655419409

As mentioned in slack, Trino JAVA version has changed a lot since initial Trino Oreilly Guide 1st edition.
- trinodb.slack.com/archives/C01TEP0HJTH/p1670555267443549

As the book requires Java 17 LTS in page 27, JVM configs suggested in the book for debugging purpose also should be updated since those options has been removed before Java 17 due to bugs or etc.

I checked two sites below and run them in docker environment to see if they worked.
- docs.oracle.com/en/java/javase/17/docs/specs/man/java.html#convert-gc-logging-flags-to-xlog
- bugs.openjdk.org/browse/JDK-8145092?focusedCommentId=13877021&page=com.atlassian.jira.plugin.system.issuetabpanels%3Acomment-tabpanel#comment-13877021

Here are changes suggested.

-XX:+PrintGCApplicationConcurrentTime
-XX:+PrintGCApplicationStoppedTime

> -Xlog:safepoint
> Note that PrintGCApplicationConcurrentTime and PrintGCApplicationStoppedTime are logged on the same tag and aren't separated in the new logging.

-XX:+PrintGCCause

> GC cause is now always logged, so remove it.

-XX:+PrintGCDateStamps

> Date stamps are logged by the framework, so remove it

-XX:+PrintGCTimeStamps

> Time stamps are logged by the framework, so remove it

-XX:+PrintGCDetails

> -Xlog:gc*

-XX:+PrintReferenceGC

> -Xlog:gc+ref*=debug

-XX:+PrintClassHistogramAfterFullGC
-XX:+PrintClassHistogramBeforeFullGC

> -XX:+PrintClassHistogram
> Previous option has been removed, so suggest similar option; Enables printing of a class instance histogram after one of the following events.

-XX:PrintFLSStatistics=2

> Option has been removed.

-XX:+PrintAdaptiveSizePolicy

> -Xlog:gc+ergo\*=level
> Use a level of debug for most of the information, or a level of trace for all of what was logged for PrintAdaptiveSizePolicy.

-XX:+PrintSafepointStatistics
-XX:PrintSafepointStatisticsCount=1

> Option has been removed.

Trino server uses Java 17 LTS and for Trino server that uses a higher memory value and queries that consumes large memories, there could be issue with G1GC, especially with preventive GC.
- trinodb.slack.com/archives/CGB0QHWSW/p1670574061721919

Before calling full GC, G1GC tries young GC in separate thread to free up space. If all regions are used (old, humongous), JNI blocks GC. GC retries max twice, and throws OOM afterward, and does not call full GC.

This feature of preventive GC was added in Java 17 and turned on as default value. However, default value for the option is turned off in Java 20 and is now debating whether to remove it completely.

Therefore, I suggest guide should include on how to handle if such issue happens, by disabling preventive GC; add JVM option below:
-XX:+UnlockDiagnosticVMOptions
-XX:-G1UsePreventiveGC