* Remove unused function mkdir()
* Remove some outdated comments
* Add include guards to AgencyStrings.h
* Move check for initialized agency out of registerAtAgencyPhase1
* Whitespace cleanup
* Address two minor comments from review
* Consolidated _servers and _serverAdvertisedEndpoints, added rebootId, prepared change notifications
* Cleanup
* Added a RebootId type
* Began implementing RebootTracker (still WIP)
* Moved RebootId operators into the class
* Removed RebootId operator<< again
* Added tests, added CallbackGuard, removed/commented old RebootTracker code
* Fix: do not try to call unset callbacks
* Split one test, added another
* Added more tests
* Renamed tests, added more tests
* Fixed missing variable declarations
* Let MockServer appear to be started
* Reorded test, fixed naming
* Implemented callMeOnChange()
* Re-implemented RebootTracker (not yet working)
* Resolved a TODO, updated a test, added comments
* Call old callbacks immediately
* Fixed tests
* Use EXPECT_* instead of ASSERT_*
* Suppress a log message
* Resolved TODOs
* Reverted changes on reading ServersRegistered
* Update RebootTracker
* Introduce `rebootId` into ServerState for Cluster
* A server *boots* if it is started on a previously non-existing data
directory and hence does not have a UUID yet.
* A server *reboots* if it is started on a pre-existing data directory
We keep the rebootId in the cluster's agency under
Current/ServersKnown/$uuid/rebootId.
When rebooting (and subsequently re-joining a cluster), the server increments
its rebootId in Phase 2 of registration. This way it can be detected within the
cluster whether a server was restarted.
This information will later be used to handle cases where server restarts can
lead to problems, for example with transactions or in-progress queries.
* Move rebootId into Current/ServersKnown/
* Fixed typo
* Fixed log ids
* Add deletion of ServersKnown/UUID from agency
* Add deletion of Current/ServersKnown/UUID to removeServer
* Clean up readRebootIdFromAgency and add retry loop around it
* Bugfix
* Added nolint comments
* Fixed initialization order
* Fixed ClusterInfo-test
* Added log messages
* Revert "Fixed ClusterInfo-test"
This reverts commit d983596979.
* Disabled assertion for google tests
* Ignore windows compile warning
* Always call loadServers in loadCurrent
* Fix really subtle bug when not returning a value
* Introduce `rebootId` into ServerState for Cluster
* A server *boots* if it is started on a previously non-existing data
directory and hence does not have a UUID yet.
* A server *reboots* if it is started on a pre-existing data directory
We keep the rebootId in the cluster's agency under
Current/ServersKnown/$uuid/rebootId.
When rebooting (and subsequently re-joining a cluster), the server increments
its rebootId in Phase 2 of registration. This way it can be detected within the
cluster whether a server was restarted.
This information will later be used to handle cases where server restarts can
lead to problems, for example with transactions or in-progress queries.
* Move rebootId into Current/ServersKnown/
* Add deletion of ServersKnown/UUID from agency
* Add deletion of Current/ServersKnown/UUID to removeServer
* Clean up readRebootIdFromAgency and add retry loop around it
* Fixed compile error due to forbidden implicit cast
* Fixed compile error on windows
* Fixed compile error due to devel merge
* Removed dead comment
* Removed TODO note
* Extended comment
* Removed TODO note
* Fixed using an invalidated iterator
* Copy string only if necessary
* Fixed compile error
* Coordinators do not unregister at every shutdown.
Instead they create a new short name with every start.
This is needed for the transactions.
* Always new short id for coordinators. Never for DBServers!
* Fixed bug where the Foxxmaster doesn't reset jobs after a crash when it should, or a non-master coordinator removes jobs in progress during startup
* Added a regression test
* Updated CHANGELOG
* Fixed non-maintainer compile
- Schmutz now called "Maintenance" and completely implemented in C++
- Fix index locking bug in mmfiles
- Fix a bug in mmfiles with silent option and repsert
- Slightly increase supervision okperiod and graceperiod
* add "DBSERVER" as an alias for "PRIMARY"
This allows specifying the value "DBSERVER" for `--cluster.my-role`.
"DBSERVER" is only treated as an alias for "PRIMARY", because several
other parts of the code and APIs use the string "PRIMARY".
Changing these from "PRIMARY" to "DBSERVER" would make the change
downwards-incompatible, which we do not want.
The downside of this alias-only solution is that even when specifying
a role value of "DBSERVER", the server will still report its role as
"PRIMARY", which may be a bit confusing. The server will also generate
its id as "PRMR-XXXX" as before:
2018-08-03T15:23:09Z [9584] INFO {cluster} Starting up with role PRIMARY
2018-08-03T15:23:09Z [9584] INFO {cluster} Cluster feature is turned on. Agency version: {"server":"arango","version":"3.4.devel","license":"enterprise"}, Agency endpoints: http+tcp://[::]:4001, server id: 'PRMR-f655b728-4cea-44ac-88e9-8b34baa80958', internal address: tcp://[::1]:8629, role: PRIMARY
* adjusted documentation to use "DBSERVER" instead of "PRIMARY"
* api doc
- secondary role not used anymore. stated.
- primary database is not clear. replaced with dbserver
- brief referenced only dbserver and coordinator - better to provide wider description, in line with what is described below, as other roles can be returned
* typo
* typo
* added starting from 3.4
* additional warning
* cited in the release note
* Only update Plan and Current from Agency if not already done.
* Add read protection for getPlanVersion and getCurrentVersion.
* Add a further check to loadPlan and loadCurrent.
* Fix tests to new behaviour.
* Try to increase Plan/Version and Current/Version with every change.
* Add two more increments of Plan/Version
* Add missing increments in tests for Plan/Version.
* Add changelog entry.
* fix some deadlocks found by evil lock manager (tm)
* fix duplicate lock
* fix indentation
* ensure proper lock dependencies
* fix lock acquisition
* removed useless comment
* do not lock twice
* create either a V8 transaction context or a standalone transaction context, depending on if we are called from within V8 or not
* AQL micro optimizations
* use explicit constructor
* only use V8DealerFeature's ConditionLocker for acquiring a free V8 context
entering and exiting the selected context is then done later on without having to hold the ConditionLocker
* remove some recursive locks
* Disable custom deadlock detection when Thread Sanitizer is enabled
* Changing ifdef's
* grr
* broke gcc
* Using atomic for ApplicationServer::_server
* fix premature unlock
* add some asserts
* honor collection locking in cluster
* yet one more lock fix
* removed assertion
* some more bugfixes
* Fixing assert
(cherry picked from commit 1155df173bfb67303077fbe04ee8d909517bfd21)
* Switching from ttl to supervision based failover mechanism
* Allowing canceling of ongoing actions
* refactored asyncjobmanager
* refactoring some code
* adding read-only flag
* Fixing some bugs
* Fixing tests
* Canceling ongoing operations
* removing some unused code + some asserts
* catching some exceptions to reduce log pollution, removing unnecessary code, removing tests for _changeMode
* fixing "createsANewDatabaseWithAnInvalidUser"
* Current work
* proper ifs
* Migrate resthandler to c++ and implement mode
* readonly server mode spec test
* Add changelog
* change code so it expects a full object and not just a string
* Fix jslint
* remove some non-unused V8 persistents
* do not throw that many bogus assertions
* do not rely on server role being defined
* slightly better debug output for V8 context debugging
* fix collection ids in inventory response
* simplify bootstrap a bit
* slightly better error handling
* make elapsed time a queryable value
* use less memory for stub collections
* added assertions that will always make sense
* added assertions
* do not garbage-collect while waiting
* less copying of parameters
* do not show "load indexes into memory" buttons for mmfiles engine
as all indexes are in memory anyway
* when a collection is truncated via the web interface, flush the WAL and rotate all active journals
this will make close all open journals on leader and followers and make them subject to compaction opportunities
* fix invalid server id values being passed from web interface to backend
* introduce afterTruncate method for indexes
* added test case for issue #3447
* updated CHANGELOG
* don't warn about replicationFactor for system collections
* check that the queries actually use the geo index and not some other index
* properly report error in web interface
* fix some internals checks that made truncate fail for bigger collections in maintainer mode
* also run a compact() operation after a serious truncate
in order to make iteration over the truncated range much faster
when the collection is next accessed
* increase default maximum number of V8 contexts to at least 16
* Switching from ttl to supervision based failover mechanism
* Allowing canceling of ongoing actions
* refactored asyncjobmanager
* refactoring some code
* adding read-only flag
* catching some exceptions to reduce log pollution, removing unnecessary code, removing tests for _changeMode
* fixing "createsANewDatabaseWithAnInvalidUser"
* auth = off does not longer make everyone superuser
* Fixing cluster_sync and maybe resilience
Jan Christoph Uhde
Simon
Tobias Gödderz
Dan Larkin-York
Markus Pfeiffer
Jan
Lars Maier
Max Neunhöffer
Tobias Gödderz
Frank Celler
Wilfried Goesgens
Dan Larkin-York
Kaveh Vahedipour
jsteemann
m0ppers