1. Release Notes for 3.7.5

Platforms for Version 3.7.5

GemStone/S 64 Bit version 3.7.5 is supported on the following platforms:

• Red Hat-compatible Linux 7.9, 8.10, and 9.6, Ubuntu 22.04 and 24.04, and Debian 12 and 13, on x86_64.
  GemStone is tested on a mixture of Red Hat, CentOS, and Rocky; these are all considered fully certified platforms. Any reference to Red Hat applies to any Red Hat-compatible distribution.
• Ubuntu 24.04 on ARM. Linux ARM is for development only, not for production.
• macOS 15.5 (Sequoia) and 13.7.6 (Ventura), on x86 Apple silicon (ARM) and x86.
  macOS distributions are for development only, not for production.

For more information and detailed requirements for each supported platforms, please refer to the GemStone/S 64 Bit v3.7.5 Installation Guide for that platform.

GemBuilder for Smalltalk (GBS) Versions

The following versions of GBS are supported with GemStone/S 64 Bit version 3.7.5:

VSD Version

VSD 5.6.5 is included with the GemStone distribution, and can also be downloaded as a separate product from https://gemtalksystems.com/vsd.

GemBuilder for Java

You must use GemBuilder for Java (GBJ) v3.2.1 with GemStone/S 64 Bit v3.7.5. GBJ v3.2.1 adds support for Windows clients, but has no other functional changes from v3.2.

There there were a number of significant infrastructure changes between GBJ v3.1.3 and v3.2; see the Release Notes for GemBuilder for Java v3.2 for more information on upgrading.

Rowan and Jadeite

Rowan is an open-source source code management system for GemStone, with underlying git repositories managing tonel-format code source files. GemStone/S 64 Bit v3.7.5 includes Rowan v2.10 for legacy users and v3.5 for use with Jadeite for Pharo.

Jadeite is an open-source GUI application for developing and debugging Smalltalk code in a GemStone repository; currently alpha-level and suitable for testing and evaluation. Jadeite supports code management operations on rowanized source code in Rowan projects and packages. Jadeite for Pharo clients are installed into Pharo 13.0 images and can be used on Linux, Windows, and Mac. Issues with the Jadeite alpha should be reported as tickets on the Jadeite project, github.com/GemTalk/JadeiteForPharo/issues

For information and help on Jadeite and getting started, see the online help: docs.gemtalksystems.com/current/JadeiteHelp.

To find out more about the current status of these open-source projects, inquire on one of GemTalk’s email forums: the customer forum at gemstone-smalltalk@lists.gemtalksystems.com, or the GLASS forum at glass@lists.gemtalksystems.com.

Updated shared library versions

The version of zlib has been updated from 1.3 to 1.3.1

The version of openssl has been updated from 3.0.16 to openssl 3.0.19

Installation using APT packages

GemStone is now supported on Debian 12 and 13.

Image Changes

Log rotation process

The application sets up a handler for the LogRotateNotification signal (using addDefaultHandler:). This handler sends reopen to the file being rotated. If more than one file is rotated, even if the rotation is on different schedules, the handler should reopen all of these files. A reopen is a no-op if the file was not moved or deleted.

At the OS level, log rotation of application GsFiles should move the disk file to an archive location, and send SIGHUP to the Gem.

Note that using mv within the local file system ensures no writes can be lost. If a file write occurs within the small gap between log archiving and sending SIGHUP, there is a risk a write may be lost if the archive process does a log copy and deletes the old log, including mv to a non-local file system destination.

LogRotateNotification enableSignalling.

LogRotateNotification addDefaultHandler: [:ex | 

	myGsFile reopen ifNil: [

		self error: 'reopen failed ', myGsFile pathName].

	myGsLog reopen.

	ex resume.

	].

GcReclaim threads for tranlog restore

The method Repository >> restoreFromBackup:* shuts down the reclaimGem, and when it completes, restarts the reclaimGem with an increased number of threads, to ensure reclaim does not hold up tranlog restore. Previously, the system computed the number of threads, which could be considerably larger and did not respect STN_MAX_GC_RECLAIM_SESSIONS. Now, the maximum number of threads is the configured STN_MAX_GC_RECLAIM_SESSIONS or 4, whichever is larger.

You should review your configuration of STN_MAX_GC_RECLAIM_SESSIONS, to ensure the setting is appropriate for your application. This is most important for repositories performing a large volume of tranlog restore.

Tranlog restore does not continue if session terminated

In previous releases, when the session that initiated restore from tranlogs was terminated, the restore continued until all specified tranlogs were restored. Now, terminating the session will stop tranlog restore at that point. Restore can be continued by logging in and re-executing the tranlog restore operation.

One-step restore and commitRestore

After a programmatic restoreFromBackup* (of a repository in full tranlogging mode), you must login again to execute commitRestore. In cases where you do not need to restore tranlogs, the restore process can be simplified with the added method disableRestoreFromLogs. After executing this method (in both encrypted as well as non encrypted extent systems), the restoreFromBackup skips restoring tranlogs and automatically performs a commitRestore.

Note that any commits that are not in the programmatic backup are lost, since transaction logs are not replayed, and transaction logs cannot be replayed after commitRestore.

This method should be executed after login and before the restore operation. The status is set in SessionTemps, and only applies for that restore operation (since restore terminates the session).

Repository >> disableRestoreFromLogs
Causes the next restoreFromBackup operation (normal or secure) by the session to immediately commit the restore and return the repository to normal mode. Executing the method #commitRestore is unnecessary.

Bug Fixes related to Restore

ParallelDo Requirements

• The ParallelDoManager must run on the Stone’s host.
• ParallelDoManager must be run as a UserProfile that is not SystemUser. This UserProfile must have #CreateOnetimePassword and #SessionControl privileges, as well as privileges such as #CodeCompile that may be required to perform the processing block. The workers will run with this same UserProfile.
• GEMSTONE_NRS_ALL needs to be set to specify the Netldi name and port number (unless using default-named gs64ldi). This must be defined in the environment for the Stone and Netldi, and in the Netldi’s environment on remote hosts.
• The Netldi conforming to GEMSTONE_NRS_ALL must be running the Stone’s host and all remote hosts.

Using ParallelDo to process objects

[:obj | obj migrate ]

(ParallelDoManager

	newForCollection: myColl

	workerBlock: [:obj | obj migrate ]) 		makeCurrent.

ParallelDoManager >> addLocalHostWithNumSessions: numSessions

ParallelDoManager >> addHostWithName: hostname withNumSessions: numSessions

parallelDoMgr 	

	addLocalHostWithNumSessions: 2;

	addHostWithName: 'lark' withNumSessions: 4.

printCommitConflictDetails:

gemStatmonitorArgs:

tempObjCacheSizeMb:

tempObjCacheSizeGb:

sharedCacheSizeGb:

sharedCacheSizeMb:

sharedCacheLargeMemoryPagePolicy: 

cacheWarmerArgs:

sharedCacheLargeMemoryPageSizeMb: 

aHost := ParallelDoHost 

	newForHostname: 'lark' withNumberOfSessions: 4. 

aHost 

sharedCacheSizeGb: 4 ; 

tempObjCacheSizeMb: 200 ; 

sharedCacheLargeMemoryPagePolicy: 2 ; 

sharedCacheLargeMemoryPageSizeMb: 1024 ; 

cacheWarmerArgs: '-d -n 2 -L /tmp' ; 

printCommitConflictDetails: 2; 

gemStatmonitorArgs: '-i1 -u 3 -z -f /tmp/statmon1s_%%S_%%P_%d-%m-%y-%H:%M:%S'. 

parallelDoMgr addHost: aHost. 

Example

The following example uses two workers on localhost and two workers on a remote host, including configuration of the remote cache, to migrate objects in the collection theColl.

| parallelDoMgr aHost |

parallelDoMgr := ParallelDoManager 

	newForCollection: theColl

	workerBlock: [:obj | obj migrate].

parallelDoMgr makeCurrent.

parallelDoMgr 	addLocalHostWithNumSessions: 2.

aHost := ParallelDoHost 

	newForHostname: 'lark' withNumberOfSessions: 2.

aHost 

sharedCacheSizeGb: 4; 

tempObjCacheSizeMb: 200. 

parallelDoMgr addHost: aHost. 

ParallelDoManager current 

	createChunks;

	prepareToRun;

	run;

	pollUntilCompleteAtInterval: 500;

	collectErrors;

	createWorkerReport.

The final call to createWorkerReport prints a report formatted in columns:

GsTsExternalSession improved handling of String/Unicode string results

When code executed by a GsTsExternalSession returns a kind of string, the external session automatically converts the result into a string in the calling session.

Previously, only remote objects of type String or Unicode7 were correctly converted, since these hold Characters requiring no more than 8 bits. The behavior for these objects is unchanged; the local object created for the remote object is of the same class as the remote object.

Remote objects of DoubleByteString, QuadByteString, Unicode16 and Unicode32 were previously returned as ByteArrays. Now, these values are automatically encoded into instance of Utf8 prior to returning. This is the same process as the recommended workaround in previous releases.

Instance of Utf8 were, and are, automatically converted to a local object of the appropriate class:

• When Repository is not in Unicode Comparison Mode, a traditional string: String, DoubleByteString, or QuadByteString.
• When Repository is in Unicode Comparison Mode, a Unicode string class: Unicode7, Unicode16, and Unicode32).

Note that when the remote execution’s return values are traditional or unicode strings that do not match the default classes for the local repository’s Unicode Comparison Mode (most commonly if the remote repository has a different Unicode Comparison Mode than the local repository), the class of the result in the local repository may vary if some require encoding and others do not.

If you previously were not using the workaround, but doing the multi-step process required to decode the ByteArray into a kind of string, this will no longer work; you will need to remove the manual decode.

Note that legacy GsExternalSession has no change in behavior.

GsTsExternalSession bugs fixed

GsTsExternalSession >> flushReleaseOops

Detached Execution changes and bug fixes

Detached execution refers to RPC Gems that are started using GsTsExternalSession forkAndDetach* methods, or the newly added GsHostProcess >> forkAndDetach (GsHostProcess >> forkAndDetach).

DateAndTimeANSI now abstract

DateAndTimeANSI was conceptually intended to be an dialect-neutral superclass containing the ANSI-specified DateAndTime behavior, while the DateAndTime implementation itself was GemStone-specific extensions and implementation details. However, both were previously concrete, and instances of DateAndTimeANSI could be created. With the introduction of SmallDateAndTime, issues were introduced on some code paths. DateAndTimeANSI is now abstract.

Creating a DateAndTime

The following instance creation method has been added:

DateAndTime class >> posixSeconds: secondsSince1970 offsetSeconds: offsetSeconds
Return a SmallDateAndTime if arguments are in range. aNumber is UTC time in seconds since the beginning of 2001, anOffset is offset from UTC.

Converting between DateAndTime and DateTime

The following methods have been added, allowing conversion between the ANSI DateAndTime class and the legacy DateTime class.

DateTime instances know their specific TimeZone, but DateAndTime instances only know their offset, which could correspond to multiple different TimeZones. Two conversion methods are provided so you can either use the current TimeZone or provide a specific TimeZone as an argument for the converted DateTime instance.

DateAndTime >> asDateTime
Answer a DateTime representing the same moment in time as the receiver, in the current TimeZone. The offset of the receiver is ignored. DateTimes know their TimeZone, and the TimeZone knows about any daylight saving time transitions, but DateAndTimes only know their offset. Most offsets can occur in several different TimeZones, so we cannot infer the TimeZone from the offset.

DateAndTime >> asDateTimeInTimeZone: aTimeZone
Answer a DateTime representing the same moment in time as the receiver, in the given TimeZone. The offset of the receiver is ignored.

DateTime >> asDateAndTime
Answer a DateAndTime representing the same moment in time as the receiver, and with the offset appropriate to the receiver's TimeZone as of that moment.

Extracting Date from DateAndTime

The method DateAndTime >> asDate has been added, to extract the Date from a DateAndTime without needing the expensive conversion to a DateTime.

Change in handling of DateAndTime creation on the cusp of daylight saving time

When daylight saving time causes the clock to be set forward or back. there is a period of time in which there is either no local time, or two ambiguous local times. This was previously handled similarly to the legacy non-ANSI DateTime; this has been changed so DateAndTime conforms to the ANSI spec.

When the time is set back at the end of daylight saving time, there two local times corresponding to a single UTC time. ANSI specifies that an ambiguous local time should resolve to the earlier of the two local times. Previously, ambiguous DateAndTimes were resolved for the later of the two local times; now, the former one is created, to conform to ANSI.

When creating a DateAndTime from a local time that does not exist, previously this was allowed and created DateAndTime one hour later; for example, creating a DateAndTime for the non-existent time of 2:30 while daylight saving time is starting, created the actual local time of 3:30. The ANSI standard specifies that this should error, and attempting to create this nonexistent local DateAndTime now signals an error.

Note that the older, non-ANSI DateTime class resolves ambiguous times to the earlier of the two, and handles nonexistant times by creating the equivalant later time; DateTime is unchanged, for historic reasons. If your application requires rigorous handling of date-time values during DST transitions, it is recommended that you use UTC times, which are never ambiguous.

Incorrect values in DateTime instance creation out-of-range error

When the second or hour argument to DateAndTimeANSI class>>year:day:hour:minute:second:offset: was out of range, the error message provided an incorrect upper limit (off by one) (#51523)

DateAndTime creation with negative partial offsets handled incorrectly

DateTimes in TimeZones with a partial negative offset, that is, in a TimeZone east of UTC and west of the International Date Line, and with a offset that is not an increment of a whole hour, produced DateTimes with an offset that was incorrect by one hour. (#51657)

Unreliable methods disabled

The following methods could produce incorrect results, and the functionality has been removed; they will now return an error. The methods remain to provide the replacement method, and send deprecated: for deprecation tracking although they are not technically deprecated.

DateAndTimeANSI class >> secondsLocal:offset:
Use secondsUTC:offset: with the local time converted to UTC.

TimeZoneInfo >> offsetAtLocal:
Use offsetAtUTC: with the local time converted to UTC.

TimeZoneInfo >> transitionAtLocal:
Use transitionAtUTC: with the local time converted to UTC.

TimeZoneTransition >> transitionTimeLocal
Convert the result of transtionTimeUTC to local time.

Added methods related to Time Zone transitions

TimeZoneTransition >> asDateAndTime
Answer the transition moment expressed as a DateAndTime with the offset being transitioned to.

TimeZoneInfo >> transitionsAtOffsetPosixSeconds: seconds
Given a local 'wall clock' time encoded as the seconds relative to Jan 1, 1970 in the time zone defined by the receiver, answer a collection of transitions which would result in that time. Normally the result is size 1, but may 0 or 2 as a result of DST.

Finding reference paths to classes

Scans to find a reference path to an object, using GsSingleRefPathFinder or the recently added methods findReferencePath and findReferencePathString, now can be used to find paths to instances of Class or Metaclass3.

The GsSingleRefPathFinder printToLog instance variable is now a SmallInteger rather than a Boolean, with the following values:

0 – no printing

1 – print start and end times

2 – also print start and end times of each scan

3 – also print limit set details for each scan

PrivateObject >> findReferencePath 

PrivateObject >> findReferencePathString 

Added method to find all instances with a single reference

A new method has been added to find all objects in the repository that have only a single unique reference.

Repository >> listObjectsWithOneReferenceWithMaxThreads: maxThreads waitForLock: lockWaitTime percentCpuActiveLimit: percentCpu
Scans the entire repository for objects referenced once and only once by any other object, and returns these as a GsBitmap.

Multiple references from a parent object to a child object are considered a single references and the child object will be included in the result. Objects which reference themselves will be included in the result if that reference is the only one.

This method begins a transaction and runs in this transaction for its duration. It should therefore not be used in production systems due to commit record backlogs which may cause excessive repository growth.

Uncommitted objects and dead not reclaimed objects are excluded from the result. Note that objects in the result set may be disconnected from the repository (unreachable) and therefore could disappear after the next garbage collection cycle.

Raises an error if the session has modified persistent objects. Raises an error if a garbage collection operation is in progress with the repository vote state is not zero (see System class >> voteState). The lockWaitTime argument is used to specify how many seconds method should wait while attempting to acquire the gcLock. No other garbage collection operations may be started or in progress while this method is running. There also must be no outstanding possible dead objects in the system for the GC lock to be granted.

Starts maxThreads on the host system and allows the host to run up to percentCpu percent CPU usage. A page buffer of 16 pages (256 KB) is allocated per thread.

Added method to search GsBitmap by class

The following method has been added:

GsBitmap >> allCommittedInstancesOf: aClass
Searches the receiver for all committed objects which are instances of aClass, and returns a new GsBitmap containing those instances. Uncommitted objects present in the receiver are ignored, and not included in the result.

Configuring the AdminGem

Starting in v3.7.2, the AdminGem only runs when needed: after markForCollection to handle voting, and if epoch GC is enabled.

The configuration method System class >> setAdminConfig:toValue:, which transiently sets values in the currently running AdminGem, usually had no effect if Epoch was not enabled. Now, this method returns an error if invoked when the AdminGem is not running.

Use the method System class >> setPersitentAdminConfig:toValue:, to define the configuration for the current AdminGem and which will be used the next time it starts up.

Method comments relating to the AdminGem have been improved; and some minor issues in GcGem session management have been fixed.

The following methods have been added:

System class >> getPersistentAdminConfig: configSymbol
Return the AdminGem persistent configuration settings, which are values in GcUser's UserGlobals. May differ from current runtime settings. See getAdminConfig: for relevant symbols.

System class >> getPersistentReclaimConfig: configSymbol
Return the ReclaimGem's persistent configuration settings, which are values in GcUser's UserGlobals. May differ from current runtime settings. See getReclaimConfig: for relevant symbols.

Added System class informational methods

2 GcUser reclaimgcgem 1642585 on localhost

3 SymbolUser symbolgem 1642586 on localhost

4 DataCurator topaz -l 1705674 on localhost transactionLevel 1, this session

5 SystemUser gem 1693847 on localhost transactionLevel 1 not voted

6 GcUser admingcgem 1705766 on localhost

userId DataCurator, transactionLevel 1, viewTime 28 seconds ago, onOldestCr true, commitsSinceView 0, 

gemProcessId 3814914, gemHost localhost, gciPeer 127.0.0.1, login 28 seconds ago, tempOops 1999

ClassOrganizer support for class categories

Methods have been added to provide information about class categories. These methods accept and return kinds of String, since they are designed to support GUI tools. These category methods treat hyphen-delimited class category strings as representing a hierarchal structure and the added methods are provided to query both for specific full category strings and association with logically-inherited supercategories.

ClassOrganizer >> classCategoryNames 

ClassOrganizer >> allClassCategoryNames 

ClassOrganizer >> classCategoryNamesInDictionaryName: dictionaryName 

ClassOrganizer >> allClassCategoryNamesInDictionaryName: dictionaryName 

ClassOrganizer >> classNamesInClassCategoryNamed: categoryString dictionaryName: dictionaryName 

ClassOrganizer >> classNamesUnderClassCategoryNamed: categoryString dictionaryName: dictionaryName 

See the image method comments for more details.

Other added image methods

Rowan-related image changes and bug fix

Un-deprecated method

The method System class >> myUserGlobals has been un-deprecated. UserGlobals has been a required SymbolDictionary for several major releases. It is used by the new ParallelDo feature.

DateTime removed/deprecated methods

DateAndTime >> currentTimeZone

DateAndTime class >> secondsSince2001

DateAndTimeANSI >> initializeAsNow

DateAndTimeANSI class >> secondsLocal:offset:

TimeZoneInfo >> offsetAtLocal:

TimeZoneInfo >> transitionAtLocal:

TimeZoneTransition >> transitionTimeLocal

DateAndTime class >> _scaledDecimal6mantissa:

DateAndTime class >> _smallFromMicrosecs:offset:

DateAndTimeANSI >> asFloatParts

DateAndTimeANSI >> partsFrom:

DateAndTimeANSI >> printString (now inherited)

DateAndTimeANSI >> _posixSeconds:offset:

DateAndTimeANSI >> _seconds:

DateAndTimeANSI >> _secondsLocal:offset:

DateAndTimeANSI class >> _zoneOffsetAtLocal:

DateAndTimeANSI class >> _zoneOffsetAtUTC:

DateAndTimeANSI >> asDays

DateAndTimeANSI >> asPosixSeconds

DateAndTimeANSI >> asString

DateAndTimeANSI >> beRounded (now shouldNotImplement) 

DateAndTimeANSI >> currentTimeZone

DateAndTimeANSI >> posixSeconds:

DateAndTimeANSI >> printJsonOn:

DateAndTimeANSI >> printRoundedOn:

DateAndTimeANSI >> printStringWithRoundedSeconds

DateAndTimeANSI >> rounded

DateAndTimeANSI >> _secondsUTC:offsetSeconds:

DateAndTimeANSI class >> fromString:

DateAndTimeANSI class >> migrateNew

DateAndTimeANSI class >> posixSeconds:offset:XX

DateAndTimeANSI class >> secondsUTC:offset:

DateAndTimeANSI class >> _zoneOffsetAtPosix:

Other removed methods

The following methods have been removed:

GsSingleRefPathFinder >> printTimestampToLog

ProcessorScheduler >> dbgfatallog:

Repository >> _getShrinkRepository

Repository >> _primRestoreSecureBackups:scavPercentFree:bufSize:privateDecryptionKey:passphrase:numThreads:shrinkRepos:newSystemUserPassword:

Repository >> _restoreBackups:scavPercentFree:bufSize:numThreads:shrinkRepos:newSystemUserPassword:

System class >> _cacheStatWithName:opCode:

Split tranlogs using startlogsender have changes in file permissions

When startlogsender is invoked to write split tranlogs (using the -F and -W flags), the split tranlogs are created using the default umask & 666, and each one is made read-only when closed. Previously, the mask was 400, removing group and world read; now, the previous file mode & 0444 is applied, allowing group (or world) read depending on the default umask.

New utility script to validate configuration files

Previously, the only way to detect errors in configuration files was to attempt to start the Stone, or login, using that configuration file.

Now, you can validate a configuration file for use for the Stone, Gem, or X-509 secured Netldi. Note that this checks for syntax errors and required settings, but is not designed to determine that a particular set of extents can start up using that configuration file, e.g. for missing extents or available RAM.

os$ validate_config -t 1 -p $GEMSTONE/data/system.conf

os$ validate_config --type=1 --path=$GEMSTONE/data/system.conf

largememorypages and SHR_PAGE_CACHE_NUM_PROCS

The largememorypages calculation relies on an accurate value for SHR_PAGE_CACHE_NUM_PROCS for the machine that will be configured with Linux Huge Pages.

SHR_PAGE_CACHE_NUM_PROCS is normally left at a default value in configuration files. When this is at the default, GemStone calculates a value based on a number of individual configuration parameters, including STN_MAX_SESSIONS, but also including, for example, the number of reclaim threads and the number of cachewarmer threads. These parameters also support defaults that allow GemStone to calculate a value based on the environment, including the number of cores.

If SHR_PAGE_CACHE_NUM_PROCS is not explicitly set, and the -P argument is not used, the results are particularly inaccurate for remote cache calculation using largememorypages -r.

For calculations for the Stone’s host:

• For the most accurate results for a specific Stone and host, run largememorypages with the specific configuration file to be used and on the actual host that the Stone will be running on, setup exactly as it will be used.
• For reproducible results, such as while modifying the configuration or running on a different host, run largememorypages with the -P argument or an explicit setting for SHR_PAGE_CACHE_NUM_PROCS

New shared library to support statprom utility

The statprom utility provides an interface to use Prometheus to monitor GemStone by recording statistics from the GemStone shared page cache. In this version, statprom code is packaged in a new shared library, libstatprom-3.7.5-64.so. This simplifies installation in some configurations.

There are no changes in behavior or configuration.

Topaz Changes

   error.topaz

   simple.topaz

   template.topaz

Examples

startstone -l %S-%T-%%Y-%%m-%%d-%%H:%%M:%%S.log 

gs64stone-stone-2026-01-07-14:44.log

gs64stone_390702reclaimgcgem.log

gs64stone_390703symbolgem.log

gs64stone_390671pcmon.log

gs64stone_login_2026-01-07-14:44:08.974.log

EXPORT GEMSTONE_CHILD_LOG_PATTERN=GS_%%Y-%%m-%%d-%%H:%%M:%%S_%T_%P.log

startstone -l GS_%S_%%Y-%%m-%%d.log 

GS_gs64stone_2026-01-07.log

GS_2026-01-07-14:25:14_pcmon_389149.log

GS_2026-01-07-14:25:14_reclaimgcgem_389180.log

GS_2026-01-07-14:25:14_symbolgem_389181.log

GS_2026-01-07-14:25:14_login_389147.log

startnetldi -g -l $GEMSTONE/logs/%T%Q.log -D $GEMSTONE/logs/  34567

set gemnetid !#netldi:34567#log:%N_%%Y%%m%%d-%%H:%%M:%%S.%%q_%P.log!gemnetobject

$GEMSTONE/logs/netldi34567.log

$GEMSTONE/logs/gemnetobject_20260107-13:46:39.238_385695.log

Remote cache reconnect over unreliable network

This release includes improved handling of remote caches over unreliable networks. (#51593). Note that 3.7.5 also includes the fix for bug #51839 which was present in the 3.6.9 LD release.

A new configuration parameter has been added to automatically attempt a reconnect after a remote cache connection timeout, rather than terminating the remote cache and the sessions on that cache.

Statmonitor data file location for GEM_STATMONITOR_ARGS

When a linked login on a remote node triggers remote cache monitoring due to the configuration parameter GEM_STATMONITOR_ARGS, and the -f argument was a relative file path, the statmonitor data file was put into the home directory of the Unix user of the linked process.

Now, the data file is written to the working directory of the linked process that triggered the remote cache creation.

Removed Configuration Parameter

The obsolete configuration parameter GEM_TEMPOBJ_START_ADDR has been removed.

NetLDI reports number and duration of PAM calls

New arguments have been added to startnetldi to enable reporting of the number and duration of calls to PAM for authentication, and to trigger more detailed reporting if the duration takes more than a configured duration threshold. Reports are not printed if there has been no activity during the preceding interval.

This feature is only valid when the NetLDI is authenticating the user via PAM, that is, not in guest mode (-g); and not for the X509-secured Netldi (-S).

The added startnetldi arguments are:

-y printIntervalSecs
Interval in seconds to print statistics on calls to PAM to the log file. Not compatible with -g or -S. When the -d flag (to enable debugging ) is used and -y is not specified, a default of -y300 is inferred.

-z printThresholdMs
Print detailed statistics on calls to PAM when any PAM call takes longer than printThresholdMa milliseconds. Not compatible with -g or -S.

--- 10/06/25 12:10:13 PDT 

PAM Performance: Calls total: 55, ok: 52, fail: 3, Time avg: 8 ms, high: 103 ms, low: 0 ms, Threads: 1

--- 10/06/25 12:18:27 PDT PAM call exceeded threshold of 5 ms

Duration: 17 ms, success: 1, user: 'gsadmin'

Function         Duration calls ok fail retry err

=================================================

getpwnam         0        1     1  0    0     0  

pam_start        1        1     1  0    0     0  

pam_authenticate 14       1     1  0    0     0  

pam_end          0        1     1  0    0     0  

=================================================

--- 10/06/25 12:29:25 PDT PAM call exceeded threshold of 5 ms

Duration: 6 ms, success: 0, user: 'gsadmin'

err msg: 'Password validation failed for user gsadmin, pam_authenticate error:7, Authentication failure'

Function         Duration calls ok fail retry err

=================================================

getpwnam         0        1     1  0    0     0  

pam_start        2        1     1  0    0     0  

pam_authenticate 4        1     0  1    0     7  

pam_end          0        0     0  0    0     0  

=================================================

Netldi added ability to record statistics for VSD

New arguments have been added to startnetldi to allow it to record statistics to a vsd-compatible statistics data file. This process does not use statmonitor, and does not attach to the shared page cache; the statistics are written directly to disk by the Netldi.

The interface to manage writing statistics is basic; most of the many statmonitor features have not been implemented for the Netldi. In particular, note that statistics data will be written to a single file for the lifetime of the Netldi. To manage these statistics data files, you will need to regularly restart the Netldi, providing a new -f filename argument.

The Netldi-generated statistics data files includes Netldi and OS system statistics only. The Netldi internal stats (see Added Netldi-specific statistics) are not accessible using statmonitor or from Smalltalk methods.

-f statFileNameOrDir
A file name or directory for writing statistics data. Requires -j. Also enables writing a PAM summary (-y) on Netldi shutdown only.

If a file name is given, the file must not exist. Files with a .gz suffix will be gzip compressed; files with a .lz4 suffix will be lz4 compressed. Otherwise file is not compressed.

If a directory is given, it must exist; with a directory argument, a file with a default file name containing the Netldi name and a timestamp with a .gz suffix will be created.

-j intervalSecs
Sample interval in seconds to write statistics data to the statistics data file. Requires -f.

Added Netldi-specific statistics

The following statistics are collected only when the startnetldi cache statistics feature is enabled; these are not taken from the shared cache and are not accessible using statmonitor or from Smalltalk methods.

In addition to these statistics, the standard OS statistics for the netldid process and Linux system statistics are also recorded by the Netldi.

ClientRequestAvgTime
Average time in microseconds that a client request takes to process.

ClientRequestCount
Total number of client requests processed by netldi.

ClientRequestFailCount
Number of failed client requests processed by netldi.

ClientRequestOkCount
Number of successful client requests processed by netldi.

ClientRequestTotalTime
Total amount of time in milliseconds netldi has spent processing client requests.

ClientThreadsActiveCount
Number of client threads active in the netldi.

ClientThreadsConfigCount
Maximum number of client threads the netldi is configured for.

ClientThreadsWaitAvgTime
Average time in microseconds spent waiting for a free client thread.

ClientThreadsWaitCount
Number of times a request waited for a free client thread.

ClientThreadsWaitTotalTime
Total amount of time in milliseconds spent waiting for a free client thread.

ForkAvgTime
Average amount of time in microseconds spent in a call to fork.

ForkCount
Total number of forks performed by netldi.

ForkFailedCount
Number of failed forks performed by netldi.

ForkOkCount
Number of successful forks performed by netldi.

ForkTotalTime
Total amount of time in milliseconds spent in calls to fork.

PamCallsAvgTime
Average duration in microseconds all calls to PAM.

PamCallsCount
Number of calls to PAM

PamCallsFailedCount
Number of failed calls to PAM

PamCallsFastestTime
Duration in milliseconds of the fastest call to PAM.

PamCallsOkCount
Number of successful calls to PAM

PamCallsSlowestTime
Duration in milliseconds of the slowest call to PAM.

Netldi statistics

The Netldi now supports collecting cache statistics, and a number of Netldi-specific statistics have been added. See Netldi added ability to record statistics for VSD.

Aliases for GcAdmin sessions stats offsets were incorrect

Aliases allow meaningful names to be attached to SessionStat00-SessionStat48. The aliases had become incorrect for the GcAdmin gem, so the displayed meaningful names were incorrect. Note this fix is in VSD v5.6.5 that is bundled with the GemStone distribution, not in the GemStone/S server itself.

New stat for total Gem memory use

GemMemoryFootPrintKb (Gem)
Approximate total memory footprint of allocated temp obj memory.

Added and removed statistics for huge memory pages

Previously, Linux system statistics for HugePages did not differentiate between different memory page sizes. These old statistics have been replaced by new ones that include the page size in the name.

The following statistics have been removed:

HugePagesTotalKB

HugePagesFreeKB

HugePagesRsvdKB

HugePagesSurpKB

HugePageSizeKB

The following statistics have been added. Note that the unit of these statistics is pages, not KB/MB.

HugePages2MbTotal
The total number of 2 Mb huge memory pages configured.

HugePages2MbFree
The number of free 2 Mb huge memory pages.

HugePages2MbRsvd
The number of reserved 2 Mb huge memory pages.

HugePages2MbSurp
The number of surplus 2 Mb huge memory pages.

HugePages1GbTotal
The total number of 1 Gb huge memory pages configured.

HugePages1GbFree
The number of free 1 Gb huge memory pages.

HugePages1GbRsvd
The number of reserved 1 Gb huge memory pages.

HugePages1GbSurp
The number of surplus 1 Gb huge memory pages.

Linux ARM supports, in addition to 2MB and 1 GB huge pages, the additional huge memory page sizes 64 KB and 32 MB. The following are reported on Linux/ARM only:

HugePages64KbTotal
The total number of 64 Kb huge memory pages configured.

HugePages64KbFree
The number of free 64 Kb huge memory pages.

HugePages64KbRsvd
The number of reserved 64 Kb huge memory pages.

HugePages64KbSurp
The number of surplus 64 Kb huge memory pages.

HugePages32MbTotal
The total number of 32 Mb huge memory pages configured.

HugePages32MbFree
The number of free 32 Mb huge memory pages.

HugePages32MbRsvd
The number of reserved 32 Mb huge memory pages.

HugePages32MbSurp
The number of surplus 32 Mb huge memory pages.

Added Errors

The error corresponding to LogRotationNotification has been added:

RT_ERR_SIGHUP/6026 

An error type has been added to allow better error reporting for loss of connection during a multithreaded operation.

FATAL_ERR_DURING_MT_OP 4011

The following errors, related to internal errors in X509-secured GemStone, have been added:

ERR_REBUILD_SCAVENGABLE/4022

ERR_DEPMAP_FAILURE/4023

ERR_OT_AUGMENT_FAILURE/4024

ERR_COMPOSE_CR_FAILURE/4025

Removed Errors

The long-obsolete error GS_ERR_SHRPC_INVALID_CONFIG/4011 has been removed.

startHostAgent log file specification

Previously, the HostAgent log file was always written in the directory specified by the startnetldi -D argument, with the name hostagent-stoneName-remoteHost-PIDStoneHost.log.

The startHostAgent script now includes the -l argument. allowing the name and location of the log file to be specified. The patterns described New utility script to validate configuration files can be used.

File name precedence:

1. The -l argument to starthostagent

2. GEMSTONE_NRS_ALL in the environment where starthostagent is executed.

3. GEMSTONE_NRS_ALL in the environment where startnetldi was executed, or the startnetldi -X argument NRS

4. The default, hostagent-stoneName-remoteHost-PIDStoneHost.log

Reconnect script for mid-cache HostAgent

On a X509-secured mid-level cache, if the HostAgent that is running on that mid-level cache goes down, but the cache itself is still running, it is possible for the HostAgent to be restarted and reconnect.

To simplify this, a reconnect script is now automatically generated by the Netldi when a cache becomes a mid-level cache, with the name restartMidHostAgent_stoneName.sh. This is a no-argument script that must be manually executed.

This script is written to the directory specified by the startnetldi -D argument.

X509 mid-level cache startup now fails without required configuration

Now, if the NETLDI_PORT_RANGE and other required configuration parameters are not supplied, the mid-level cache in an X509 configuration will not start up. (#51552)

GsX509Parameters now support quietLogin flag

The following methods have been added:

GemStoneX509Parameters >> clearQuietLogin 

GemStoneX509Parameters >> quietLoginFlag 

X509-Secured GemStone bug fixes