1. GemStone/S 64 Bit 3.3.1 Release Notes

GBS version 8.1

GBS version 5.4.3

For more details on supported GBS and client Smalltalk platforms and requirements, see the GemBuilder for Smalltalk Installation Guide for that version of GBS.

Upgrade process and results

The ICU version varies between the supported originating repositories for upgrade to 3.3.1, and 2.x and 3.0.x did not have the ICU feature. The upgrade scripts and Installation Guide instructions have been updated for each upgrade path.

The result of following these instructions for each version is as follows:

• Upgrade from 2.x or 3.0.x will use ICU v54.1.
• Upgrade from 3.1.x will use ICU v54.1; you must resort/rebuild for ICU collation, since 3.1.x used an older ICU version that is not supported for 3.3.x.
• Upgrade from 3.2.x will use ICU v51.2, you must edit the configuration file setting for GS_LIB_ICU_VERSION, but no resort/rebuild for ICU collation is required.
• Upgrade from 3.3 will use ICU v54.1. If you have previously upgraded to 3.3 from 3.2.x, and you have not already rebuilt/resorted for ICU collation, you are exposed to bug #46056 and will need to perform the rebuild/resorts.

GsDevKit_home systems that are upgraded using the upgradeStone script do not have to take any further action; this script performs all necessary steps on all supported originating versions.

Distribution changes

Both 51.2 and 54.1 ICU libraries are included in the distribution, and the library interfaces are designed so that Gems in version 3.2.14 or later 3.2.x versions, or version 3.3.1, may load either version.

Image changes

A class variable and access method has been added to IcuCollator to request the version of the ICU libraries that are currently loaded:

IcuCollator class >> libraryVersion

A new entry has been added to Globals to store the version of the ICU libraries that the repository should be using:

Globals at: #IcuLibraryVersion

This is a string that is set to '54.1' or '51.2' on upgrade, depending on what version you are upgrading from.

Library load changes

The new configuration parameter STN_GEM_LIBICU_VERSION has been added, and is used by the Gem during login to determine which version of libicu*.so should be loaded. This configuration parameter is not usable on AIX due to the way libraries are loaded, and you must use the following environment variable.

The default setting for this is 54.1. An additional manual step is required on upgrade from 3.2.x to set this configuration parameter to 51.2 in the Stone’s configuration file.

The new GS_LIB_ICU_VERSION environment variable can override the setting for the configuration parameter. This may be set to either ’51.2’ or ’54.1’.

During login, if the loaded ICU library does not match the repository’s configuration, if
IcuCollator libraryVersion ~= (Globals at: #IcuLibraryVersion)), then an error is signalled and further commits are disallowed by this session. This avoids any risk of committing incorrectly sorted results.

The version test on login can be bypassed for upgrades, if the user is SystemUser and the environment variable GS_LIB_ICU_BYPASS is defined.

After subsequent upgrades, the repository will automatically continue to use the same ICU version; you must manually upgrade to use a later ICU version.

Added Configuration Parameter

Other changes related to sorting and index-related changes

The method SortedCollection class >> resortInstancesFromFilesForGem: gemNum of: totalGems has been added. This method reads the files in $upgradeLogDir/scbm that contain OOPs of instances of SortedCollection, and performs a resort.

Code has been added to IndexManager to support rebuild of indexes that rely on ICU library collation.

This code is designed for use in future improved index management during upgrade and ICU version upgrade.

ExecBlock recompile

The following methods have been added:

needsRecompile
Returns true if the receiver's block needs recompilation, false otherwise.

recompile
Attempts to recompile the receiver's method, assuming it is a simple block without references to self or method variables.

These methods test if block recompile is needed, and perform the simple block recompiles. They are invoked by postconv after a GemStone server upgrade that requires code recompile.

Complex blocks that have references to self or method variables cannot be automatically compiled, since the referents are in other methods. Handling such blocks requires individual attention; you may have to manually construct code that recreates the block.

sortWithBlock:* methods accepting IcuCollator

The sortWithBlock:* methods now provide variants with an added collator: argument. This allows you to pass in the IcuCollator as a third argument in a sort block. This allows IcuSortedCollection sortBlocks to be simple blocks, which can be automatically recompiled.

The class SortBlockUnicodeNode has been added to support this functionality. This is a subclass of SortBlockNode, used by BlockSorter to implement the sort used by sortWithBlock:*.

Change in errors

Error number 2230 was previously LGC_ERR_PACKET_KIND_BAD; this is now LGC_ERR_SERVER_PACKET_KIND.

Error number 2231 was previously LGC_ERR_EXPECTED_CONTINUE; this is now LGC_ERR_CLIENT_PACKET_KIND.

topaz output pushnew limit increased

Previously, up to 100 files could be specified; now, 1000 are allowed.

startnetldi added option to skip reverse DNS lookups

The following option has been added to startnetldi:

-N  Use numeric IP addresses for printing peer info in logs, do not do reverse DNS lookups to get hostnames of clients.

printlogs utility allows suppression of displaying strings

The output of the printlogs utility is used for debugging and for analysis of repository transaction. When printing page information, the output can include data incidental to the actual issue being examined, in the form of strings that may include sensitive information.

Now, the added nostrings argument to printlogs suppresses printing the contents of all kinds of strings and similar byte collections, including ByteArrays.

gslist "exe deleted" status

When a GemStone distribution is replaced while netldi, stone, or cache are running, gslist described these as killed; now it reports a new status "exe deleted".

GciPerformFetchBytes

GciPerformFetchBytes is similar to GciPerform, with the addition of maxResultSize, the maximum number of bytes to be returned in *result . The function returns the number of bytes returned in *result, or -1 if an error is available to be fetched with GciErr.

(int64) GciPerformFetchBytes(

OopType receiver,

const char *selector,

const OopType *args,

int numArgs,

ByteType *result,

ssize_t maxResultSize

);

GciNbPerformFetchBytes

GciNbPerformFetchBytes is a nonblocking variant of GciPerformFetchBytes.

(void) GciNbPerformFetchBytes(

OopType receiver,

const char *selector,

const OopType *args,

int numArgs,

ByteType *result,

ssize_t maxResultSize

);

GCI compilation flag

Compile flags in gcicmn.h now include GCI_COMPILE_evaluateWithoutSelf = 4, which disallows references to self, used to recompile ExecBlocks.

ICU version change caused problems in upgrade to 3.3

The ICU library version was updated for v3.3. This created issues with indexes and SortedCollections that used the ICU libraries for collation. (#46056).

The way ICU versioning is handle is changed substantially in this release, see Handling of ICU shared libraries.

Upgrade to 3.3 did not handle SortedCollection recompile requirement

After upgrade to v3.3, all code requires recompile. This includes code in SortedCollection sortBlocks and other persistent blocks. The upgrade instructions did not include this requirement. (#46055)

The Installation Guides have been updated for v3.3.1, and the code that supports postConv now recompiles simple sortBlocks from 3.2.x.

Upgrade did not correctly handle epochGcPageBufferSize

When upgrading from v2.x to 3.x, the setting for epochGcPageBufferSize was not being converted from a previous setting to a power of two. (#46009)

upgradeSeasideImage failed as a result of recently added class

After additions of GLASS/Seaside/GsDevKit classes GsFailedMethodCompilationDuringCopyMethodToNewClass, and NumberParserGsDevKitIssue75Test, the upgradeSeasideImage step failed. (#46059, #46217)

Corrupt object error during symbol conversion from 2.4.x

During the symbol conversion phase of the 2.4.x to 3.x conversion, a data page could end up with an out of range clusterId. This resulted in a corrupt object error. (#46032)

DbfHistory extra line for 3.2.x-originated repositories

In repositories originating in 3.2 and later, the DbfHistory included lines both for a 3.2.0 filein and a filein specific to the actual originating version. (#45977)

Class instance referenced from class prevented GC

If a application class’s class variable referenced an instance of the class, but there are no other references to the class or any instance of the class, and the size of the instance is 0, the class may have been immune from garbage collection. (#46116)

Gem crash due to excessive recursion in MFC

There was an optimization in the multithreaded MFC that performed a recursion; it was possible for excessive recursion to occur which resulted in running out of Cstack space. This optimization was determined to provide very little value and has been removed. (#46194)

Reclaim Gem may startup in transaction

When the stone starts up and has to recover due to an unclean shutdown, and there were pages to reclaim, then the Reclaim Gem started up in transaction. This could cause a commit record backlog until recovery completed, or Reclaim Gems’ maxTransactionDuration was reached. (#46005)

More information in cases where Admin Gem is not running

If the Admin Gem is not running following an operation that holds the GC lock, a request for a markForCollection cannot get the gcLock and fails to start with an error. The error did not indicate the cause of the problem; now, the error message includes a warning that the Admin Gem is not running. (#46047)

Repeated replay of transaction log can cause corruption

If an incomplete transaction log was replayed, and then later the same transaction log was replayed after it had completed (and therefore contained more records), it was possible for some records in the later part of the transaction log to not be replayed. This resulted in the restored system being corrupted. (#46148)

Possible deadlock in multithreaded backup after file error

If a file read or write error occurred while a multithreaded backup is being written, it attempted to write the error to the log without releasing a mutex, resulting in a deadlock. (#46243)

.gz not appended to short backup file names

When creating a compressed backup using fullBackupCompressedTo:, a ".gz" should be automatically appended if it is not provided in the filename string. If the filename string was fewer than 4 characters, the ".gz" suffix was not appended. (#46169)

GsQuery reject: incorrect results

The code for GsQuery reject: included a bug and returned the wrong results. (#46184)

GsQuery detect:ifNone: returned query instead of ifNone: result

Rather than returning the results of the ifNone: block, on a negative result it returned the query. (#46184)

ReversedRangeIndexReadStream size incorrect

Sending #size to an indexing reverse read stream returned 0, rather than the number of elements in the results to be streamed over. (#46184)

Index Audit could miss invalid encryption keys

Index audit did not catch all cases of invalid encryption keys in interior nodes. (#46069)

Remote shared page cache monitor can get stuck spin lock

There was a code path in which a remote shared page cache could have deadlocked on a stuck spin lock. (#46175)

After remote cache startup timeout, cache status may be stuck

If a remote cache fails to complete startup within the time period of STN_REMOTE_CACHE_STARTUP_TIMEOUT, the stone starts shutting down and cleaning up after the startup attempt. There were timing cases in which the remote cache could complete the startup and shut itself down, leaving the Stone status incorrect. (#46020)

Page Manager crash on removing pages from remote cache

It was possible for the Page Manager thread in the stone to detect a fatal error and crash, when there was a mismatch in the number of pages that it had requested to remove from remote shared page caches. (#46171).

Configuration Parameter Changes

the maximum for GEM_TEMPOBJ_MESPACE_SIZE and GEM_TEMPOBJ_POMGEN_SIZE has been increased from 1000000 to 64000000

The minimum for GEM_TEMPOBJ_MESPACE_SIZE has been changed from 1000 to 2000.