What’s new in dCache 9.2
Release notes
Highlights
- Performance improvements for the concurrent directory creation and removal
Acknolagements
Many Thanks to Lars Jansse and Sandro Grizzo for their contributions.
Incompatibilities
- prior to version 9.2 dCache NFSv4.1 door will publish only
nfs4_1_files
layout. Now on the door publishes all available layout types - dropped reference count tracking for directory tags
- dropped support of java options
chimera_soft_update
andchimera_lazy_wcc
in favor ofchimera.attr-consistency
property - If upgrading from 8.2, be sure to read also the release notes for 9.0 and 9.1, important changes are described there. See the 9.2 migration guide for more complete information
- With version 9.2.8+, empty and non-existent banfiles will be treated the same
Release 9.2.29
tape
Users reported 2 day pin lifetime on staged files (which is a default) despite specifying different values. This is now fixed.
Changelog 9.2.28..9.2.29
- 0d911615bc
- [maven-release-plugin] prepare release 9.2.29
- ab9a74538d
- tape REST api: additional fix tohandling of prefixed paths
- 499959a572
- [maven-release-plugin] prepare for next development iteration
Release 9.2.28
xroot
Return destination address (that is, the haproxy address) if xrootd.enable.proxy-protocol=true
is set, instead of the actual door address.
Changelog 9.2.27..9.2.28
- 57579ad5de
- [maven-release-plugin] prepare release 9.2.28
- 95cad94b15
- xroot: handle haproxy and checksum command
- 89133016c8
- [maven-release-plugin] prepare for next development iteration
Release 9.2.27
CI
Pipeline optimizations.
pool
Fix double decrement on active hsm requests. This addresses the issue of pools stopping flushing to tape with “Negative number of active requests”.
When a thread performing I/O gets interrupted, then an InterruptedIOException might be thrown. A DCAP mover will treat and propagate such an exception as a disk I/O error, thus disabling the pool. This fix reduces false positive disk I/O errors.
Changelog 9.2.26..9.2.27
- c2f32fc913
- [maven-release-plugin] prepare release 9.2.27
- d8d5541241
- pool: don’t treat InterruptedIOException as a disk IO error
- 8e376b7181
- pool: fix double decrement of hsm requests
- 1215548076
- ci: split container image registry and repository
- 0863dacffa
- [maven-release-plugin] prepare for next development iteration
Release 9.2.26
bulk
The current release fixed broken command activities
.
Changelog 9.2.25..9.2.26
- 9e9ea4fa6d
- [maven-release-plugin] prepare release 9.2.26
- 6664395575
- ci: run spotbugs only on master branch
- b075c04d19
- bulk: fix broken command
activities
- 9c0b160e42
- [maven-release-plugin] prepare for next development iteration
Release 9.2.25
bulk
When specifying an empty target, bulk proceeded to process the request instead of failing fast. This is now fixed.
gplazma
A previous commit, leading to the last bugfix releases being blacklisted, introduced a regression in the multimap plugin. Where the ‘op’ principal type is used, logins will fail with dCache logging a stacktrace like
java.lang.RuntimeException: Failed to create principal: java.lang.NoSuchMethodException: org.dcache.auth.OAuthProviderPrincipal.<init>(java.lang.String)
.
This is now fixed.
Changelog 9.2.24..9.2.25
- 8fecee871f
- [maven-release-plugin] prepare release 9.2.25
- 4d14890f7d
- gplazma: fix broken commit d74d9568167f4
- 729ce8f76a
- gplazma: multimap fix op regression
- 5708994ab6
- bulk: check targets for empty strings
- 6043a20af4
- [maven-release-plugin] prepare for next development iteration
Release 9.2.24
doc
Better documentation clarifying OIDC provider ID and issue claim.
webdav
Bumps org.eclipse.jetty:jetty-servlets from 9.4.51.v20230217 to 9.4.52.v20230823. Thi has fixed java.lang.NullPointerException: null error on WebDAV Domain due to jetty 9.4.51 bug.
Changelog 9.2.23..9.2.24
- bfaeae101e
- [maven-release-plugin] prepare release 9.2.24
- 6394fbecd4
- github: add action for atumatic github-release
- 6ab84e51ea
- gplazma alise initial version of plugin
- db6ea0f464
- common: add issuer URI to OAuthProviderPrincipal
- 09e5a49937
- build(deps): bump org.eclipse.jetty:jetty-servlets
- eb858c9725
- docs: clarify OIDC provider ID and issue claim
- 7e6098c30e
- [maven-release-plugin] prepare for next development iteration
Release 9.2.23
CI
Improve our CI pipleline.
core
Pool migration : fixed behavior of migration copy
with
-replicas=n
(where n>1
) to behave as expected.
Libs
Improve and maintain our testing library.
Keeping libraries up to date.
TAPE-API
Release by relative path works.
Changelog 9.2.22..9.2.23
- 63803072d4
- [maven-release-plugin] prepare release 9.2.23
- a7aa1b3689
- WLG TAPE REST API: fix handling of frontend.root in release API
- a219e4457a
- gitlab: mirror tags
- 3de9e9301c
- libs: update mina-sshd to version 2.13.1
- 96ef3777ca
- Fix issue with infinite replicas when replicas > 1
- 6b3ff705aa
- github: change mirroring action
- 7bd322edc4
- ci: use shorter k8s namespace names
- abe0f45324
- pom: add exta java option to run powermock test under java17
- 118a5cc34d
- ci: sync CI pipeline with master
- fb954fb8ef
- ci: use desy nims repo for CentOS7
- 562668fc1e
- ci: use almalinux9 for rpm install test
- 2991f5ef12
- [maven-release-plugin] prepare for next development iteration
Release 9.2.22
chimera
Now it is possible to n read/write CTA and migrated Enstore files without having to set hsmInstance tag in the full directory tree.
frontend
The current release fixed the issue about failures to invoke WLCG tape API to stage files on files relative to {webdav,frontend}.root
It is now possible to call stage API successully on paths ralative to frontend.root.
Changelog 9.2.21..9.2.22
- b7f34147e4
- [maven-release-plugin] prepare release 9.2.22
- 743c52948e
- pom: compile code in the same jvm as maven
- 8a428d3316
- chimera: add CTA HSM StorageInfo extractor
- 69f2f109ca
- frontend: handle frontend.root variable properly
- 5652f98acf
- [maven-release-plugin] prepare for next development iteration
Release 9.2.21
chimera
Adapt existing uri_encode
function so that the Enstore HSM script does not fail and the files go to tape successfully.
ci
CI pipeline improvements.
dependencies
Update build dependency to be compatible with modern tools.
Changelog 9.2.20..9.2.21
- f48df60d4e
- [maven-release-plugin] prepare release 9.2.21
- 9691e0096c
- dependencies: update modernizer plugin to be compatible with maven 3.9
- 6e2c1186bd
- chimera: fix uri_encode to handle special characters
- f52a0ee2f4
- ci: add property to control upload options
- 24759f2e3c
- [maven-release-plugin] prepare for next development iteration
Release 9.2.20
bulk
Bulk truncates path to 256 characters and this seems to be causing problemsd.
This is now fixed.
pool
With CentOS7, the client sends the last request, such that offset+count == filesize. REHL9 sends the last request so that the count is multiple of 4096, which is legal.
The current release fixed miscalculation of offset on short read.
Changelog 9.2.19..9.2.20
- f3b6d8e7c9
- [maven-release-plugin] prepare release 9.2.20
- 8d7d4661d2
- ci: disable srmcp test
- a710037a84
- ci: use python3 for robot test
- 4158c80ae5
- bulk: do not truncate target paths
- 9b9714f267
- ci: pinpoint postgres helm version
- 15ea025a3f
- pool: fix miscalculation of offset on short read.
- 39bb2f532d
- [maven-release-plugin] prepare for next development iteration
Release 9.2.19
chimera
Fixed a bug that made it possible to create a loop on directory move.
Changelog 9.2.18..9.2.19
- 368ce200f4
- [maven-release-plugin] prepare release 9.2.19
- 465e00e82a
- chimera: fix loop creation on directory move
- 9a46793677
- ci: drop –ftp-create-dirs option (as we switch to https)
- c0357f4f81
- [maven-release-plugin] prepare for next development iteration
Release 9.2.18
webdav
When running multiple Remote Transfermanager, two transfer that were started simultaneously had the same ID which led to the second transfer becoming orphan after the first one completed. This is fixed now.
Changelog 9.2.17..9.2.18
- 9cb77bc505
- [maven-release-plugin] prepare release 9.2.18
- b0e6eaf759
- ci: eplocitly specify kubernetes namespace
- 3f06a7a6ad
- ci: fix typo
- 5cf1c757c9
- webdav: use transfermanager+id to identify TPC transfer
- 73fbb8b77f
- docs: update oidc chapter to explain trust anchors
- 25fe810ba9
- [maven-release-plugin] prepare for next development iteration
Release 9.2.17
common
The out-of-box version of ProxyCSRGenerator#generate from CAnL uses SHA1 for proxy delegation, which is banned by modern OSes.
this now fixed and RHEL9 clients works with proxy delegation without enabling SHA1.
xroot
Switch to xrootd4j–4.6.0, ew major release with bug fixes and enhancements including the reload TLS certificate.
Changelog 9.2.16..9.2.17
- 42faeb1884
- [maven-release-plugin] prepare release 9.2.17
- 973a06353d
- xroot: switch to xrootd4j–4.6.0
- 3eeefe01c2
- common-security: add custom version of ProxyCSRGenerator#generate
- 177d75f696
- [maven-release-plugin] prepare for next development iteration
Release 9.2.16
qos
Admins now have the option to disable role based authorization for QoS transitions.
webdav
dCache now supports HTTP-TPC transfers where the dCache-local path includes HTTP reserved characters.
Changelog 9.2.15..9.2.16
- c4035193dd
- [maven-release-plugin] prepare release 9.2.16
- 4f8f2099ed
- qos: add flag to enable/disable role based authorization for transitions
- 92c2350067
- webdav: httptpc percent-decode local path
- e08dc2a171
- [maven-release-plugin] prepare for next development iteration
Release 9.2.15
bulk
Some of the format strings that manage the formatting of command results were sending to the user expect a different number of parameters than provided. This is now fixed and the returned bulk info strings return all expected values.
info
DGAs send messages to other dCache cells to discover their current status. These messages have a hard-coded one second timeout.
Some queries are data-intensive and could take longer than one second to build the answer, resulting in no information being provided.
this is now fixed and the info service will now wait longer for a cell to respond to a query for information.
nfs
Grizzly memory management uses heap memory size fraction even if direct memory is used (issue 7529). This issue has been fixed and dCache starts with 256m of direct memory.
Changelog 9.2.14..9.2.15
- 04a4d03830
- [maven-release-plugin] prepare release 9.2.15
- 253e07ab7d
- info: use DGA refresh rate as message timeout
- 58cc1f30f5
- nfs: calculate desired memory fraction for correct memory allocation
- f51fcde6db
- bulk: fix divergent command params format strings
- cfa39a1960
- [maven-release-plugin] prepare for next development iteration
Release 9.2.14
cells
The error reporting on tunnel disconnect has been fixed.
srr
The current release fixed empty path annotation
warning.
tape-api
Request supplied paths are now mapped based on frontend.root.
Changelog 9.2.13..9.2.14
- 7ff7f5ba6a
- [maven-release-plugin] prepare release 9.2.14
- f9ef94a8dd
- tape-api: map request supplied paths based on frontend.root
- 2c8a17f3ad
- srr: fix “empty path annotation” warning
- fa7cd0ec2e
- cells: fix error reporting on tunnel disconnect
- bb4b067fb9
- [maven-release-plugin] prepare for next development iteration
Release 9.2.13
core
This patch has been shown to fix the issue in which poolmanager will in some instances not load parts of its configuration.
packaging
Packaging infrastructure improvements.
Changelog 9.2.12..9.2.13
- 3d735b9766
- [maven-release-plugin] prepare release 9.2.13
- 23469021c9
- Revert “poolmanager: delete property for switchingon/off caching for psu”
- 861b5a104c
- ci: use minimal almalinux–9 to upload packages
- 453fa408a8
- [maven-release-plugin] prepare for next development iteration
Release 9.2.12
pool
Buffer initialization is now done at pool start instead of first NFS read request. This leads to more predictable buffer initialization.
Changelog 9.2.11..9.2.12
- 7ee8097efc
- [maven-release-plugin] prepare release 9.2.12
- fee2057919
- ci: use dtzar helm-kubectl image
- 481408f770
- dcache-core: fix psu logging
- 764779040b
- pool: mover grizzly IO buffer initialization into NfsTransferService
- 82cd451b32
- [maven-release-plugin] prepare for next development iteration
Release 9.2.11
webdav
Users have reported severe slow down when runnig listing using webdav.
This is currently fixed.
Changelog 9.2.10..9.2.11
- 46f47153a8
- [maven-release-plugin] prepare release 9.2.11
- da22a29c73
- webdav: fix slow listing
- 2d73d8ba98
- [maven-release-plugin] prepare for next development iteration
Release 9.2.10
webdav
A recent path introduced support for adding the Link HTTP response header, according to RFC 6249. Unfortunately, the code added the Link header for all requests, not just the intended GET and HEAD requests.
Additionally, due to peculiarities of how Milton generates PROPFIND results, the Link header is added multiple times: once for each subdirectory.
This results in the HTTP response headers taking up too much space and dCache failing the request, returning a 500 status code.
This regression has been now fixed where PROPFIND requests would fail if the directory contains too many subdirectories. The cut-off point depends on the length of the URL the PROPFIND request targets.
Changelog 9.2.9..9.2.10
- f58389b35b
- [maven-release-plugin] prepare release 9.2.10
- f15e622cb4
- webdav: fix link header
- 3125694915
- [maven-release-plugin] prepare for next development iteration
Release 9.2.9
gplazma
A bug is fixed in gPlazma that is triggered when users attempt to authenticate with broken gPlazma configuration or a gPlazma plugin’s configuration is broken.
Changelog 9.2.8..9.2.9
- ffef10d130
- [maven-release-plugin] prepare release 9.2.9
- c7993fe0f3
- gplazma: fix NPE if gPlazma is rejecting all logins
- 42d497e172
- [maven-release-plugin] prepare for next development iteration
Release 9.2.8
chimera
A recent commit as a side effect resulted in failure to upload a file to an xroot door with the overwrite (“-f”) option, returning okay but removing the original file without creating the new one. This patch reverts the offending commit: copy with an overwrite flag works again.
CI
Improve build pipeline.
gplazma
Previously, a missing banfile would result in every successive login attempt failing with an NPE. Now, even if configured, an empty or inexistent banfile will be ignored and logins should succeed. WARNING: This changes the banfile plugin behaviour! Empty and non-existent banfiles will be treated the same.
system-test
Increases the small default direct memory value for system-test: NFS on system-test works again.
webdav
Update webdav door to retry the request to the namespace if checksum is not present in files attributes.
xroot
A recent commit introduced handling of relative paths in xrootd with inadvertent side effect of dropping support xroot.root variable. This patch reverts that commit: pre–9.2 behavior is restored.
Changelog 9.2.7..9.2.8
- 63c0dfbd70
- [maven-release-plugin] prepare release 9.2.8
- 7dcd2cc2fd
- ci: dont’t pull build artefact for kubernetes-based jobs
- 1af0614d0c
- Fix unit test for commit 6b47354
- 0deb226953
- gplazma: configured banfile plugin should ignore non-existent ban file
- 2ac2180724
- chimera: Revert “chimera: update FsSqlDriver#inodeOf to throw exception if file not found”
- 38d2cae1ff
- system-test: increase direct memory
- 3840f87195
- xrootd: fix xrootd.root regression
- baa135bab9
- webdav: wait for upload to complete
- 16d3bb94a3
- [maven-release-plugin] prepare for next development iteration
Release 9.2.7
ci
Build and test system improvements.
cleaner-disk
No more occasional ConcurrentHashMap exceptions in cleaner-disk runs due to concurrent pool status changes.
core
Admin commands with hyphen-containing arguments are no longer trunkated prematurely, they are now properly shown and autocompleted.
qos
Prevent PropertySetterException
in qos.
xroot
The change to use effectiveRoot did not take into account Subject = Nobody.
Regression eliminated, previous behavior with anonymous read restored.
Fixes previous fix for xroot descriptor: NPE risk mitigated.
Changelog 9.2.6..9.2.7
- a91a3bbb1c
- [maven-release-plugin] prepare release 9.2.7
- 8a60a76165
- dcache-xroot: check that descriptor is not null before calling close – fix
- 84e5af54cd
- dcache-xroot: check that descriptor is not null before calling close
- 711b4a38f8
- cleaner-disk: prevent ConcurrentModificationException in cleaning run
- 84026b3779
- dcache-xroot: fix effective root when subject is nobody
- 95590b4a5f
- dcache-core: fix admin command completion
- e8bc2c0274
- qos: comment out property description
- c2022ea407
- ci: add exta helm ops to extend timeout and simplify retries
- 53afeee632
- [maven-release-plugin] prepare for next development iteration
Release 9.2.6
dependencies
The MongoDB driver was updated. This might lead to different log messages.
frontend
A bug was fixed that led to an Exception when using qos without policy.
webdav
If a non-existing well-known resource is requested, it won’t log a stacktrace anymore.
Changelog 9.2.5..9.2.6
- c0bc7fa627
- [maven-release-plugin] prepare release 9.2.6
- a83cc7fbfd
- pom: Update mongodb-driver
- 1b10c15afb
- dcache-frontend: check for defined arguments with namespace qos
- 1748b3b887
- webdav: do not log stacktrace if non-existing
well-known
is requested - 9ad93f688c
- [maven-release-plugin] prepare for next development iteration
Release 9.2.5
bulk
The current releae fixed handling of uncaught exception.
dcache-view
dCache-view has been updated to 2.1.0 and it includes modifications to use new RolePrincipal instead of the roles plugin role, eliminating the need for additional logins for admin privileges.
qos
With ingest queues servicing large numbers of requests, as we have for performance reasons configured as defaults on QoS and Bulk, there was a potential for a race between the processing of the original modify request and a subsequent cancellation request, such that cancellation could not find the request as it was still in the executor queue.
This is fixed now and are is a trailing stream of requests still being processed after cancellation (from Bulk) completes.
Changelog 9.2.4..9.2.5
- 7969d6bbd7
- [maven-release-plugin] prepare release 9.2.5
- dee8cbc483
- dcache-qos: cache modify requests until processed by executor
- 3a255bb92c
- dcache-bulk,dcache-qos: repair mass cancellation issues
- 4bed61af7e
- dcache-bulk: fix handling of uncaught exceptions
- 73b49ce730
- pom.xml: update dcache-view to 2.1.0
- ea69bea591
- Fixed QoSPolicyTest duration strings
- 347c668ce9
- dcache-frontend,common: check parsing of Duration in STAGE
- 470e0f65a3
- [maven-release-plugin] prepare for next development iteration
Release 9.2.4
chimera
The rollback of 9.1 to 9.0 db schema was fixed and possibility to rollback namespace db to an earlier schema version is restored now.
webdav
The urls similar to this https://door.domain.foo:1234//pnfs/domain.foo/path/to/file
were strip into /domain.foo/path/to/file
.
The current release fixed the parsing of urls with two slashes in the path.
Changelog 9.2.3..9.2.4
- cd710f3992
- [maven-release-plugin] prepare release 9.2.4
- cefea2f3cb
- chimera: fix rollback of 9.1 to 9.0 db schema
- 95fcd5d230
- webdav: fix infinite recursion in Requests.stripToPath
- 0e33db68c5
- webdav: fix parsing of urls with two slashes in the path
- 428f7c77e5
- ci: generate release-notes template
- 8fff819a39
- Revert “ci: use rancher to create k8s namespace”
- 9e4b101bcf
- [maven-release-plugin] prepare for next development iteration
Release 9.2.3
dcache-bulk
Performance and stability improved, but throughput continues when the submitted task activities are in a state of waiting for future completion.
door
Exception handling is improuved for Kafka.
oidc
Storgae scopes
without path will be rejected now.
Changelog 9.2.2..9.2.3
- b8134f8db3
- [maven-release-plugin] prepare release 9.2.3
- ec5783ca89
- dcache-bulk: use rate limiter to throttle semaphore release
- 4e9864a336
- oidc: fix remove invalid testcase
- 8059c50501
- oidc: reject storage scopes without path
- af96f5c6d6
- door,pool: handle multiple possible KafkaExceptions
- c9fe2f61f9
- docker: use exec for start java process
- 674bb383df
- [maven-release-plugin] prepare for next development iteration
Release 9.2.2
bulk
A bug was fixed that may lead to the archiver deleting ongoing requests.
qos
The default size of the task queues are set to “qos.limits.verifier.max-running-operations” now to increase the throughput. IMPORTANT: The default size of the QoS Engine’s modify thread pool was also updated to 500 as the current value, 32, is too little.
webdav
This patch fixes a bug that led to wrong paths on redirects.
Changelog 9.2.1..9.2.2
- 16ffa5d557
- [maven-release-plugin] prepare release 9.2.2
- 3278975650
- ci: use pynfs:0.5, enable LOCK24 test
- 09c872a80a
- ci: use rancher to create k8s namespace
- 38678c9209
- dcache-qos: correction to the threshold warning
- 1a8bb02376
- dcache-qos: set default task thread pool sizes all to max concurrent running
- b3a1bb7b6b
- dcache-bulk: fix bug in archiver deletion query
- ea7e26199b
- dcache-bulk: fix thread executor injection
- d466c360c0
- [maven-release-plugin] prepare for next development iteration
- 08c0ac9c60
- wevdav: fix redirect path
Release 9.2.1
bulk
Correct bulk cancellation semantics.
The problem with cancellation (hanging perpetually in the CANCELLING state) was solved.
Return directory listing to its own executor.
Same performance as before, but with a total memory footprint at a little more than half available physical memory instead of being very close to it.
Add commands to display the count instead of the actual entry (as with requests and targets), and also a command to clear/delete from the archive table.
documentation
Add commands to display the count instead of the actual entry (as with requests and targets), and also a command to clear/delete from the archive table.
Fixes link in documentation.
Add section in WebDAV door chapter on metalink.
Changelog 9.2.0..9.2.1
- 8d4bf82df5
- [maven-release-plugin] prepare release 9.2.1
- 8cac6fa53e
- dcache-bulk: fix cancellation issues
- 66b67eabe3
- dcache-bulk: give directory listing a separate executor
- 151b4cb86a
- doc: add QoS policy and role documentation to the dCache Book
- 0ffd722bd6
- docs: UserGuide fix angle-braket in metalink
- 02924858c5
- docs: add description of webdav’s metalink support
- be306f450e
- dcache-bulk: refine container executor model
- 3cd0912aca
- dcache-bulk: cancel activity future on target cancel
- 201359918b
- dcache-bulk: add count and clear to archive admin commands
- 95f53d8478
- [maven-release-plugin] prepare for next development iteration
Release 9.2.0
Known issues
Cancel is currently broken in bulk service and recursive requests could freeze up.
Admin
Paging capability was added to the AnsiTerminalCommand (and DirectCommand), where results exceeding 10K lines prompt the user with [Y/N] for further results. When executing a shell command in non-terminal mode, all results are streamed back continuously. The Bulk commands have been converted to use this feature.
Bulk
There have been many small fixes and improvements to this service since 8.2.
The most significant changes or additions include:
The storage layer has been redesigned to conserve space and for efficiency/throughput. Please note that moving up to the new database schema may require some time.
The amount of time can be generally computed in terms of the number of entries in therequest_target
table; figure about 1 hour for every 10 million. If it is necessary to maintain all such entries, we recommend you do the upgrade offline, using thedcache database update
command-line tool. If there is no need to keep completed requests in the database, we would advise truncation or at least deletion of the completed requests before the upgrade.Periodic archiving of requests has been added; this is configurable via properties and admin commands. The archive table maintains an abbreviated summary of the requests, and can be purged via admin command as well.
The container job has been rewritten to bring it in line with SRM bringonline performance; multithreaded directory listing has also been implemented to improve recursive request time-to-completion.
Path resolution (for relative paths) has been integrated into bulk request processing.
Activity providers now can capture the environment so that defaults can be customized (this presently pertains only to pin and stage lifetime attributes).
Support has been added for the QoS update request to handle policy arguments.
Bulk has been made replicable (HA). Please read the requirements in the cookbook section on High Availability.
The
delay clear
option has been eliminated from bulk requests.The
prestore
option has also been eliminated, as it is no longer necessary (since all initial targets are immediately batch stored synchronously) before the submission request returns.
A few minor points:
Default limits for request size and number of requests per user have been set on the basis of WLCG requirements.
The counts displayed by the
info
command are from startup and are cumulative; for actual (current) counts based on the data store, use thestatus counts
command.
There are numerous properties which have been deprecated, particularly those dealing
with limits. Please review the bulk.properties file for details. It should not
be necessary under normal usage to adjust the limits for thread pools, database
connections and semaphores from the defaults. Depending on the volume of activity
and the site requirements, the archiver period and window may need to be shortened
from the default. Note that the options for clearOnSuccess
and clearOnFailure
that can be included in an individual bulk request to indicate immediate deletion
are not available for STAGE requests as this feature is not part of the WLCG specification,
so the only way to remove such requests automatically is through the archiver.
Chimera
With dcache version 9.0 chimera has introduced chimera_soft_update
and chimera_lazy_wcc
Java properties to control the behavior
of the parent directory attribute update policy. Now those properties are obsolete and replaced by a regular dcache configuration
property chimera.attr-consistency
, which takes the following values:
policy | behaviour |
---|---|
strong | a creation of a filesystem object will right away update parent directory’s mtime, ctime, nlink and generation attributes |
weak | a creation of a filesystem object will eventually update (after 30 seconds) parent directory’s mtime, ctime, nlink and generation attributes. Multiple concurrent modifications to a directory are aggregated into a single attribute update. |
soft | same as weak, however, reading of directory attributes will take into account pending attribute updates. |
Read-write exported NFS doors SHOULD run with strong consistency
or soft consistency
to maintain POSIX compliance. Read-only NFS doors might
run with weak consistency
if non-up-to-date directory attributes can be tolerated, for example, when accessing existing data, or soft consistency
,
if up-to-date information is desired, typically when seeking newly arrived files through other doors.
Frontend
Aside from bug fixes, the following changes should be noted:
- Support for .well-known/security.txt was added to both the frontend and WebDav ports.
- More detailed description of request objects for bulk and stage.
- Improved error messages in several places.
- Support for relative paths and symlink prefix resolution for bulk and namespace resources.
- The frontend.wellknown paths have been deprecated in favor of dcache.wellknown.
- Authz checks have been removed in Quota GET methods.
- Use of RolePrincipal (see under gPlazma) replaces reliance on LoginAttributes and the old admin role (special gid) as defined by the roles plugin.
- Support for QoS Rule Engine policies. A new qos-policy resource allows one to add, remove, list and retrieve policy definitions. See the Swagger pages for details.
- An -optional query parameter was added to the namespace resource to retrieve extra information about a file; the new QoS Policy file attributes (QOS_POLICY and QOS_STATE) are included with this option.
- Support for pool migration has been introduced (migrations resource). See Swagger pages for details.
gplazma
A RolePrincipal has been added to support the use of role definitions in the multimap file. Currently the available roles are: admin, qos-user and qos-group. The second allows the user to transition files owned by the user’s uid; the third allows the user to transition files whose group is the user’s primary gid. The two qos roles can be combined. Admin grants full admin privileges. A multimap example:
dn:"/DC=org/DC=cilogon/C=US/O=Fermi National Accelerator Laboratory/OU=People/CN=Al Rossi/CN=UID:arossi" username:arossi uid:8773 gid:1530,true roles:admin
These roles do not depend on the presence of the roles plugin. Since the frontend has been changed to use the new RolePrincipal and not the old role definitions (where a special gid must be defined), one can easily drop that plugin from the gplamza.conf file.
There was also an important fix for a bug that was preventing upload (e.g., xroot POSC) when using tokens.
The gplazma-xacml plugin has been dropped.
NFS
Prior to version 9.2 dCache, to support RHEL6 based clients, if no species export options are specified, the NFSv4.1 door
were publishing only nfs4_1_files
layout. Now on the door publishes all available layout types. If for whatever reason
RHEL6 clients are still used, the old behavior can be enforced by lt=nfsv4_1_files
export option.
PNFS Manager
To remove unused directory tags chimera keeps the reference count (nlink) of tags. This approach creates a ‘hot’ record that serializes all updates to a given top-level tag. Starting 9.2 dCache doesn’t rely on ref count anymore and uses conditional DELETE, which should improve the concurrent directory creation/deletion rate.
Pool
The mover ls
command is updated to display with the -u
option the subject associated with the mover.
Added pool.mover.https.port.min
and pool.mover.https.port.max
to control TPC port number used by HTTPS mover.
Pools info
command displays HTTP, HTTPS, NFS and XROOT movers listen ports and interface.
Billing information from DCAP and FTP movers now on include local socket end point
Poolmanager
Pnfsid and path options have been added to ac psu match.
Re-introduce wrandom
partition type - a weighted-random pool selection partition, which works as a WASS
partition, but ignores LRU metric and number of movers.
QoS
Aside from bug fixes, the following significant changes to QoS should be noted:
QoS was made to support migration using a new pool mode, “DRAINING”; please consult the book chapter for further details.
-
The DB namespace endpoint can now be configured to be separate from the main Chimera database (originally introduced/changed for Resilience). In this way, the scanner, whose namespace queries are read-only, could be pointed at a database replica.
This remains possible for QoS, even though the QoSEngine now also is responsible for updating Chimera with file policy state. These writes are all done via messaging (PnfsHandler) rather than by direct DB connection, so they go to the master Chimera instance.
Requests for QoS transitions are now authorized on the basis of role (see under gPlazma).
-
The first version of the QoS Rule Engine has been added. With this, one can define a QoS policy to apply to files either through a directory tag or via a requested transition; the engine tracks the necessary changes in state over time. A new database table has been added to the qos database. Remember that if you are deploying QoS for the first time, you need to create the database:
createdb -U <user> qos
Fuller explanation of the rule/policy engine is forthcoming in the Book chapter.
In conformity with the new rule engine changes, the scanner has been modified in terms of how it runs scans. There are now two types of system scans, the (NEARLINE) QoS scans (to ensure that files with a policy requiring them to be flushed have indeed been written to tape) and a system-wide ONLINE scan (there are two versions of this and an option to choose which one). Details in the Book; also refer to the relevant admin commands.
Note that the singleton QoS service (where all four components are plugged into each other directly) is no longer available; the four services can, however, still be run together or in separate domains, as with any dCache cell.
Resilience
Resilience is still available in 9.2, but should be considered as superseded by the QoS services. We encourage you to switch to the latter as soon as is feasible. Remember not to run Resilience and QoS simultaneously.
A recipe has been added to the cookbook for migration/draining of resilient pools. This continues to apply in QoS.
WebDAV
The WebDAV door now provides limited support for metalink format.
Metalink is an XML format, described by RFC 5854, that describes how
to download multiple files. The initial support is limited to
describing the files within the targeted directory: there is no
support for recursion. The metalink information is available via HTTP
Content Negotiation (client sends a Accept:
application/metalink4+xml
header) or via a Link
HTTP response
header (see RFC 6249) when generating the normal HTML output.
XRootD
Aside from bug fixes and various documentation additions and clarifications, the following should be noted:
- Proxying through the xroot door is now available.
- Relative paths are supported in the xroot URL.
- Resolution of symlinks in path prefixes and paths is supported.
- The efficiency of the stat list (ls -l) has been greatly improved.
An Xroot User Guide page has been started.
Changelog from 9.1.0 to 9.2.0
- 95f53d8478
- [maven-release-plugin] prepare for next development iteration
- 86ab6a2f5d
- [maven-release-plugin] prepare release 9.2.0
- 8d0d926ab1
- gplazma: oidc update explicit AuthZ parsing
- 11ceff7a92
- gplazma: oidc increase cache duration for OP public key material
- b640b5e92c
- dcache-bulk: container rewrite to optimize threading
- b2b76e9e15
- dcache-bulk: add admin command and query to reset all requests with failed targets
- 74bab6447e
- dcache-bulk: add convenience admin command for state counts
- 12a1fbf5f4
- dcache-qos: fix scanner operation completion logic
- 28b0bebc2d
- ci: by-pass docker.io for bitnami charts
- ab16db9102
- dcache-bulk: implement HA
- 1378259897
- dcache-cli: convert IllegalArgumentException to CommandException on call()
- 2507eea428
- dcache-bulk: do not PIN or STAGE files with AL ONLINE
- a29351aae1
- dcache-bulk: only set request status to QUEUED when permissions and targets are all inserted
- 8479f44e1e
- Create test.txt
- d23083495f
- [maven-release-plugin] prepare branch @{releaseLabel}
- 707cf4f1cb
- pool: on failed upload set file size to zero for space reporting
- afe112ed48
- dcache-bulk: guard against erroneous argument names
- bf9bb6c5e3
- dcache-qos: remove entry from rule engine table if file not found or other namespace exception
- 34b411557f
- dcache-frontend: include new QoS Policy attributes in -optional
- 805735ff23
- book: remove v6.2 references in HA chapter
- 449651ff22
- poolmanager: fix wrandom partition incompatibilities
- fb04d1c990
- Revert “poolmanager: remove wrandom partition type”
- 3a33ae774c
- Add a correct check for the existance of the specified link, poolgroup or HSM when using Migration via REST.
- 0fea8a858f
- docs: gplazma document oidc plugin
- 67508e933b
- pnfsmanager: inroduce limit on number of concurrent listing of the same directory
- c1e3728412
- packages: fix system test populate sed expression
- 2a01b13aa3
- dcache-webdav: revert improve efficiency of directory listing (14085/14088)
- 29f340d3c0
- dcache-qos,system-test: override db property to hsqldb
- 8ee98957db
- ci: collect billing records from kafka
- a518359f78
- doc: clean up the multiple protocol configuration explanation for xroot
- 1266bb7372
- dcache-bulk: fix warning for premature stop of job container
- 17453da77a
- dcache (qos): move required attributes to Transfer.readNamespaceAttributesAsync
- a1140fceff
- pool: include local endpoint for ftp transfers
- effabfe737
- pool: report dcap local endpoint
- 52602da836
- ci: pin all jobs to dcache-dev runners (temporary)
- a8de5478b5
- dcache-frontend,dcache-bulk,dcache-qos: regularize observance of the admin role
- 49cf4062a9
- dcache-bulk: handle InterruptedException in DirListTask
- 41f34d8b43
- ci: trace node where gitlab agent is running
- af134b683e
- ci: use egi-trust repo
- 560dce66e4
- ci: use almalinux9-minimal images for rpm signing
- 3567dde692
- dcache-webdav: fix incorrect parameter value given to DcacheDirectoryResource constructor
- 253859a532
- dcache-qos: drop subject from qos_operation table (qos engine 9)
- 8cb60b96cd
- dcache-frontend: add support for qos policies (qos rule engine 8)
- 13580d972c
- dcache-bulk: modify qos update activity to support policies (qos rule engine 7)
- 0ebfa0553b
- dcache-qos: add policy support to scanner (qos rule engine 6)
- 710ee615e2
- dcache-qos: throttle verifier requests to engine (qos rule engine 5)
- 77bd7a1fb9
- dcache-qos: add support for policy handling in qos-engine (qos rule engine 4)
- e2a1f371eb
- dcache,pnfs: implement qos policy support (qos engine 3)
- e3d67b8c30
- chimera,dcache-chimera: add support for inode policy info (qos engine 2)
- 21af809d58
- chimera: implement qos policy storage (qos engine 1)
- 1f79ce55ce
- webdav: add metalink support
- 707000c026
- dcache-webdav: improve efficiency of directory listing
- 7e691a4f53
- dcache-frontend,dcache-webdav: use RolePrincipal instead of LoginAttributes roles
- 7609229430
- common,gplazma-multimap,dcache-qos: qos role authorization, revised
- fe522a5409
- ci: enable srm tests
- 603ab40ccc
- ci: run junit test during rpm build
- ec1ff2b5f4
- ci: add srmls tests
- 2a40be361c
- ci: report gridtest test results
- 2fb9170283
- chimera: update FsSqlDriver#inodeOf to throw exception if file not found
- c425a62257
- ci: add first robot test
- 09e8e32390
- dcache-qos: rework verifier operation handling so most of it is in memory (as in resilience)
- 8c92f2c413
- common: fix NPE in AbstractUidPrincipal#equals
- 2b83b2c135
- ci: upload_rpm must match all other upload rules
- 410762eb55
- common,gplazma-multimap,dcache-qos: authorize qos transition based on role
- de47e03e3a
- dcache-qos,dcache-bulk: allow qos update to call engine asynchronously
- 4af8e26ce0
- skel,logback: make the .resilience and .qos log files singletons
- 87dc1a4ea5
- build(deps): bump org.springframework.kafka:spring-kafka
- 6643935656
- pool: remove hidden defaults for rh/rm/sh operations
- 0dde36e585
- ci: use CI_REGISTRY_IMAGE variable to address container image
- 99bfff3c52
- dcache-bulk: adjust semaphore permits to something more reasonable
- b5e8dd5e3f
- ci: add xroot based tests
- 9e88a1ae07
- ci: better readablity and only logic based refactoring
- 841d53731c
- dcache-bulk: reconfigure activity providers to capture environment
- a32aa038a6
- rpm: add required packages
- 84c2caa61d
- docs: fix external link to nfsmapid and DNS TXT Records docs
- 4071315e36
- docs: add dcache door generic workflow
- 21f441b4ee
- Revert “ci: pass extra pgk and rpm to dcache container”
- 1398b38eef
- dcache-bulk: cancel all stored targets on abort
- ef26f40e9a
- dcache-bulk: implement periodic archiving of old completed requests (part 2)
- e36714d42a
- CI: Inheritance for Kubernetes
- d5c7c9d3c6
- ci: fix image labels
- ccb6c8d491
- ci: add labels to produced containers
- f7b8b4de3a
- ci: pass extra pgk and rpm to dcache container
- e5fdc5d741
- ci: fix manual upload logic
- 8dcb21e0a2
- ci: install required OS tools for dcache container
- 9cc8f8a376
- ci: add workaround installation of shadow-utils
- a62142e6f9
- pool: Save hsm load provider to setup file
- 02e0ef6028
- ci: use dependencies instread of command not found: needs
- 1de87b4131
- remove extra listing of workspace
- e8d7e9fc10
- ci: use pynfs–0.4 container
- b396a798ab
- ci: collect logs before cleanup
- e007ede840
- dcache-bulk: implement periodic archiving of old completed requests (part 1)
- a29d8f4f08
- dcache-xroot: improve efficiency of stat list (ls -l)
- 83e5974301
- dcache-qos: remove singleton service configuration
- d09c80074a
- dcache-bulk: remove in memory running state counts
- c0021411ec
- pom.xml: upgrade to xrootd4j 4.5.8
- ab7c3b1a4e
- ci: install shadow-utils before other packages
- 4f7f78160f
- ci: add possibility to manually publish non tagged packages
- 27d348c711
- ci: almalinux:9-minimal as based image with java17
- a565601190
- common: modify the way the flag on isRestricted works
- aa0ce5b75e
- ci: add publishing to github
- e1a450dc80
- ci: initialize worker node
- f9622f2fb2
- ci: deply test WN with shared cvmfs
- 801358fc31
- pool: show the listen port of nfs mover with admin
info
command - 2cbd89fdf9
- pool: introduce property for https mover port range
- dab3637221
- pnfsmanager: make list scheduling behavior optional (selectable)
- a848b94a1c
- chimera: rename FsSqlDriver#removeTag into removeAllTags
- 77c2dd60ce
- chimera: fix commit 8fe67b2046
- 8fe67b2046
- chimera: introduce attribute to control attribute update consistency
- f56ac72e4b
- Fixed string name in deserializer
- cdcb84ae3a
- Revert “Revert ”dcache-common: add QoS Json policy objects“”
- fa9ffb95cf
- Revert “dcache-common: add QoS Json policy objects”
- ea28525852
- dcache-common: add QoS Json policy objects
- ff857831df
- scripts: update default heap and mempry sizes
- b9dd2f410e
- ci: fix vesion of kafka helm chart (v2)
- 13de8e7a1a
- ci: fix vesion of kafka helm chart
- 7cbee56df8
- allow action only main repo
- d9ed86fcde
- docs: more pnfs flow diagrams
- 5eb4e70418
- docs: more pnfs flow diagrams
- dbd85ee88c
- docs: add pnfs flow diagram
- cdc47c7287
- github: add gitlab mirroring action
- f34a0cb4fa
- dcache-gplazma: add serialVersionUID to MultiTargetedRestriction
- ff86d14551
- chimera: drop nlink ref count for directory tags
- 2acfaf8d14
- dcache-bulk: allow multi-threaded directory listing on expandDirectories
- 33a37d6774
- fix bash-completion
- 4be34ff6c3
- pool: handle double remove of IdleStateHandler
- 633d1333ed
- nfs: expose dcache version in EXCHANGE_ID operation
- 572f76cf90
- dcache-qos: fix bug in storing and retrieving ‘tried’ value for verify operation
- 9f6910ece9
- dcache-bulk: deprecate “prestore” option and remove related container
- 5f5a5a96f8
- libs: use nfs4j–0.25.x
- 3fb66514d9
- pool: add option to display user subject with
mover ls
- c70248cc26
- ci: fix negation of disabled tests
- d767580166
- ci: disable unsupported pynfs tests
- 587ab61da1
- ci: envorce nfsv4.2 for pynfs tests
- 3c1cd449f1
- ci: use latest pynfs test set
- ae44287d91
- ci: fix exclude/include for pynfs tests
- d73359de02
- ci: set pynfs test error code according to test results
- 25f6b1ea32
- ci: fail build if pynfs tests are failing
- 17e8f603a1
- Fix Migration Resources
- 6e9e80c1da
- ci: remove obsolete kybernetes deployment script
- 2970ebaa37
- ci: use helm based dcache deployment
- ad253a69ef
- dcache-xroot: modify mkdir to ignore dir exists on make parent option
- 9e1955734a
- book: fix rpm download links
- 108f2f5a3a
- Reformat code in HsmSet.java
- 1b7e07c08e
- ci: add pynfs test into pipeline
- 88109a7d74
- ci: remove unused pipeline job
- 0edcfb12d0
- rpm: depend on java–11-headless
- b406b78bdb
- dcache-vehicles: fix NPE in resolve symlink message
- 421e1dc932
- nfs: fix race condition of LAYOUTRETURN and LAYOUTGET
- 1e54a504f9
- xrootd: throw FileNotFoundException if vomsdir doesn’t exist
- a3ca140698
- ci: deploy in kubernetes
- a9ebbc044f
- dcache-qos: fix adjuster task cloning to reset state
- 31346a795b
- dcache-qos: improve exception reported on aborted verification
- fd38a52850
- ci: prepare k8s environment for deploy and test
- 1ae82cc59c
- pool: remote http movers to provide local endpoint info
- fbdc448493
- ci: add gitlab-ci.yml
- 62e945c4e7
- docs: add DOI lable/badge
- 7f6675d7db
- gplazma-roles: add QoS role support
- ae83a08319
- dcache-http: add http request header for role assertion
- 8de990a241
- dcache-frontend: add symlink resolution to /api/v1/id GET and /api/v1/namespace
- 024b1cf5c8
- dcache-frontend: remove authz checks in Quota GET methods
- 98ad434afb
- dcache-qos: remove trigger to rescan pools on tag change
- 0d682c68de
- Revert “dcache-qos: remove trigger to rescan pools on tag change”
- 825c40d4a0
- pom.xml: bump xrootd4j to next version (4.5.7, 4.4.8, 4.3.9, 4.2.13)
- adda59a455
- dcache-qos: remove trigger to rescan pools on tag change
- ad2632019d
- several: remove non-ASCII dash occurrences
- 9eae61bbac
- dcache,pnfs: skip symlink resolution when checking restrictions on directory children during listing
- 19e5341e58
- dcache-qos: repair faulty queue refresh algorithm in verifier
- 6dde34e643
- dcache-qos(verifier,engine): fix incompatibility issues with Subject and attributes
- 34adbb0828
- skel,dcache-frontend,dcache-webdav: use dcache.well-known for all doors
- d85ec22dff
- dcache-bulk: catch Exception from getSubject()
- d8e1826672
- book: update config-message-passing.md
- 5217a1da4a
- build(deps): bump netty-handler from 4.1.92.Final to 4.1.94.Final
- 9d4c75a102
- Docs: frontend.wellknown!wlcg-tape-rest-api.path also applies to WebDAV doors
- 93ed1709f4
- [maven-release-plugin] prepare for next development iteration
- 80f485f13b
- build(deps): bump guava from 31.1-jre to 32.0.0-jre