Ganeti-htools release notes =========================== Version 0.2.7 (Thu, 07 Oct 2010) -------------------------------- Bug fixes: - fixed the error message for hail multi-evacuation mode - improve evacuation mode for offline secondary nodes (ignore available memory) New features: - add a new option ``-S`` to hbal and hspace that saves the cluster state at the end of the processing in the text format used by the ``-t`` option, for later re-processing - a two new options to hbal, -g and --min-gain-limit, that should help in limiting the number of balances steps with a low gain in the final stages - hbal, when executing jobs, will now wait for the current jobs to finish at the first stop (e.g. ^C); if the user wants immediate exit, another signal should be sent - added “normalized” physical CPU units in hspace output (NPU), which represents units of physical CPUs free/used, based on the max-cpu ratio Version 0.2.6 (Mon, 26 Jul 2010) -------------------------------- Exactly three months since the last release. Many internal changes, plus a couple of important changes in the balancing algorithm. First, the balancing may now introduce N+1 errors, if this solves other, more critical problems. For the moment, this means that moving instances away from offline nodes is allowed even if it creates N+1 errors, and that means evacuation can be done in more cases. Second, the scoring for N+1 has changed. In previous versions, it simply counted the number of failing N+1 nodes, which means moving an instance away from a N+1 failed node (but without the node 'clearing' the N+1 status) was not reflected in the cluster score. As such, the balancing algorithm managed to clear N+1 errors only sometimes, since usually it takes more than one move for this, and the first prerequisite move was not 'rewarded' appropriately and thus it was not selected. Now, it is possible to fix many more error cases than before: on a simulated 40 node cluster full with instances (symmetrically allocated on all nodes), around five nodes can be evacuated before N+1 errors can be solved, whereas 0.2.5 could evacuate at best one node. There were some other internal changes to the scoring algorithm, such that now the metrics have associated weights, and they are not all of the same importance anymore. As of now, the only change is that offline instances have a higher weight, which should favour proper node evacuations. Among the other changes: - fixed the hspace KM_POOL_* metrics, which were returned as the final state and not as the delta between the initial and final states - fixed hspace handling of N+1 failing clusters: before, it used to generate a 'fake' response, and the structure of this response was not always in sync with the real responses, leading to missing items; currently it proceeds correctly through the code (skipping the computation), and uses the same display mechanisms as the normal case - fixed hscan exit code for RAPI failures: previously it finished with success even if all the clusters failed, which was creating issues with the live-test script; now it exits with exit code 2 for RAPI failures (unfortunately this is still not optimal as LUXI failures will use exit code 1, the same as the command line) - changed the limit values for CPU/disk, which previously were used optionally, whereas now they are always used; the default cpu ratio limit is now 64 VCPUs per PCPU - changed the internal handling of the short name vs. original (Ganeti-provided) name; now internally we always use the full name, and only in display routines we show the shortened (called 'alias') name; as a result, the -O and --excluded-instances options now accept both the full name and the shortened name - changed internal handling of JSON conversions and errors, such that now we show a better context for failure messages, which should help with diagnosing the malformed message - changed the names for a few node fields, and added some more nodes; this is most likely to help with debugging, and not with regular operation though - changed the node fields option to allow the '+' prefix to mean 'extend the default fields list' rather than start from fresh (similar to Ganeti's implementation) - a few internal changes related to the LUXI protocol implementation, which should make it more safe against potential bugs, one optiomization that should help with large messages, and some patches in preparation for potential expansion of the LUXI backend functionality And finally, many improvements on unittests and the live-test script. Test coverage is much enhanced, and the test infrastructure has better error reporting; this should lead down-the-road to better code and fewer bugs… Version 0.2.5 (Mon, 26 Apr 2010) -------------------------------- Some internal cleanup plus a few user-visible changes: - new option for marking instances as 'do-not-move' during rebalancing - allow ``hscan`` to scan the local cluster via Luxi - add more metrics to ``hspace`` which show the delta between original state and final state better (only valid for tiered allocation) Version 0.2.4 (Mon, 22 Feb 2010) -------------------------------- Two improvements for node evacuation: - hbal takes a new parameter ``--evac-mode`` that restricts the instances to be moved to the ones on offline/drained nodes, which should reduce the work done - hail supports the new ``multi-evacuate`` mode of the IAllocator protocol, that will be released in a minor release on the Ganeti 2.1 branch Version 0.2.3 (Thu, 4 Feb 2010) -------------------------------- A small release: - Fixes selection of secondary node: previously, if the cluster had many N+1 failures, a N+1 failed node could be selected as secondary even if it did not have enough memory to allow the instance to be migrated/failed over to it; this is bad for automated tools, since we can get the cluster in an unhealthy state - Switch the text backend to a single input file, that is generated now by hscan and shouldn't be generated manually via gnt-node/instance list anymore; this allows richer information to be kept in the file, and simplifies a little the internals of the text backend Version 0.2.2 (Tue, 29 Dec 2009) -------------------------------- Small release, 0.2.1 was broken and thus this was released earlier: - Release 0.2.1 broke the LUXI backend due to a typo, fixed - Added a live-test script that should catch errors like the above one in the future (needs a working, non-empty cluster) - Changed RAPI and LUXI backends to treat drained nodes as offline, similar to the IAllocator backend change in 0.2.0 (which was wrongly marked as affecting all backends) - Changed the metrics for offline instances and N1 score from percent to count, in order to increase the priority of evacuations - Added a new metric (offline primary instances) which should fix the evacuation of a offline node in a 2-node cluster Version 0.2.1 (Wed, 2 Dec 2009) -------------------------------- - Added instance exclusion defined via instance tags - Fixed the output of hspace to be again parseable from the shell Version 0.2.0 (Tue, 10 Nov 2009) -------------------------------- A significant release, with a few new major features: - Added direct execution of the hbal solution when using the Luxi backend; the steps for each instance moves are submitted as a single jobs, and the different jobs are submitted as groups in order to parallelise the execution of moves - Added support for balancing based on dynamic utilisation data for instances, fed in via a text file; by default, all instances are considered equal and this change also improves the equalisation of secondary instances per node - Added support for tiered capacity calculation in hspace, where we start from a maximum instance spec and decrease the spec when we run out of resources; this should give a better measure of available capacity on 'fragmented' clusters; this is done separately from the current fixed-mode computation Also there have been many minor improvements: - Added option for showing instances (“--print-instances”), similar to the print nodes option - Added support for customising the node list via an argument to the print nodes option in the form of a comma-separated list of field names; currently the field names are not documented, expecting further changes in a next release - Enhanced the error reporting in the Luxi and Rapi backends - Changed the handling of drained nodes, now being treated the same as offline nodes, for Ganeti 2.0.4+ compatibility - A number of internal changes, simplifying code and merging some disparate functions - Simplify the build system in relation to creation of archives Version 0.1.8 (Tue, 29 Sep 2009) -------------------------------- - Brown-paper-bag release fixing haddock issues Version 0.1.7 (Mon, 28 Sep 2009) -------------------------------- - Fixed a bug in the Luxi backend for big responses - Fixed test suite exit code in presence of test failures - Changed the migrate operation to run instead failover for instances which were marked as not running in the input data (this could have been changed since then, but it's better than today's always migrate) - Added support for 'cheap' moves only (only migrate/failover) in balancing - Added support for building without curl (thus no RAPI backend) Version 0.1.6 (Wed, 19 Aug 2009) -------------------------------- - Added support for Luxi (the native Ganeti protocol) - Added support for simulated clusters (for hspace only) - Added timeouts for the RAPI backend - Fixed a few inconsistencies in the command line handling - Fixed handling of errors while loading data - The 'network' is a new dependency due to the Luxi addition Version 0.1.5 (Thu, 09 Jul 2009) -------------------------------- - Removed obsolete hn1 program; this allowed removal of a lot of supporting code - Lots of changes in hspace: the output now is a shell fragment in order for script to source it or parse it easier; added failure reasons; optimised to use less memory for large clusters - Optimized the scoring algorithm (used by all tools) so that now computations should be faster Version 0.1.4 (Tue, 16 Jun 2009) -------------------------------- - Added CPU count/ratio of virtual-to-physical CPUs to the cluster scoring methods; this means that now the balancer, the iallocator plugin and so on will try to keep the VCPU-to-PCPU ratio equal across the cluster - Fixed some hscan bugs - Fixed the way iallocator reads the total disk size (was broken and it was always falling back to summing the disk sizes) - Internals: fixed most compile-time warnings Version 0.1.3 (Fri, 05 Jun 2009) -------------------------------- - Fix a bug in the ReplacePrimary instance moves, affecting most of the tools Version 0.1.2 (Tue, 02 Jun 2009) -------------------------------- - Add a new program, “hspace”, which computes the free space on a cluster (based on a given instance spec) - Improvements in API docs and partially in the user docs - Started adding unittests Version 0.1.1 (Tue, 26 May 2009) -------------------------------- - Add a new program, “hail”, which is an iallocator plugin and can allocate/relocate instances - Experimental support for non-mirrored instances (hail supports them, hbal should no longer abort when it finds such instances and simply ignore them) - The RAPI port and/or scheme can be overriden now, and even “file://” schemes can be used if the message body has been saved under the appropriate name - Lots of code reorganization, esp. rewritten loading pipeline - Better data checking and better error messages in case validation fails; tools now consider nodes with error in input data (‘?’ returned by ganeti) as offline - Small enhancement to the makefile for simpler packaging Version 0.1.0 (Tue, 19 May 2009) -------------------------------- - Drop compatibility with Ganeti 1.2 - Add a new minimum score option (with a very low default), should help with very good clusters (but is still not optimal) - Add a --quiet option to hbal - Add support for reading offline nodes directly from the cluster Version 0.0.8 (Tue, 21 Apr 2009) -------------------------------- - hbal: prevent mismatches in wrong node names being passed to -O, by aborting in this case - add the ability to write the commands (-C) to a script via (-C<file>), so that it can be later executed directly; this has also changed the commands to include the ncessary -f flags to skip confirmations - add checks for extra argument in hbal and hn1, so that unintended errors are catched - raise the accepted “missing” memory limit to 512MB, to cover usual Xen reservations Version 0.0.7 (Mon, 23 Mar 2009) -------------------------------- - added support for offline nodes, which are not used as targets for instance relocation and if they hold instances the hbal algorithm will attempt to relocate these away - added support for offline instances, which now will no longer skew the free memory estimation of nodes; the algorithm will no longer create conditions for N+1 failures when such instances are later started - implemented a complete model of node resources, in order to prevent an unintended re-occurrence of cases like the offline instance were we miscalculate some node resource; this gives warning now in case the node reported free disk or free memory deviates by more than a set amount from the expected value - a new tool *hscan* that can generate the input text-file for the other tools by collection via RAPI - some small changes to the build system to make it more friendly; also included the generated documentation in the source archive Version 0.0.6 (Mon, 16 Mar 2009) -------------------------------- - re-factored the hbal algorithm to make it stable in the sense that it gives the same solution when restarted from the middle; barring rounding of disk/memory and incomplete reporting from Ganeti (for 1.2), it should be now feasible to rely on its output without generating moves ad infinitum - the hbal algorithm now uses two more variables: the node N+1 failures and the amount of reserved memory; the first of which tries to ‘fix’ the N+1 status, the latter tries to distribute secondaries more equally - the hbal algorithm now uses two more moves at each step: replace+failover and failover+replace (besides the original failover, replace, and failover+replace+failover) - slightly changed the build system to embed GIT version/tags into the binaries so that we know for a binary from which tree it was done, either via ‘--version’ or via “strings hbal|grep version” - changed the solution list and in general the hbal output to be more clear by default, and changed “gnt-instance failover” to “gnt-instance migrate” - added man pages for the two binaries Version 0.0.5 (Mon, 09 Mar 2009) -------------------------------- - a few small improvements for hbal (possibly undone by later changes), hbal is now quite faster - fix documentation building - allow hbal to work on non N+1 compliant clusters, but without guarantees that the end cluster will be compliant; in any case, this should give a smaller number of nodes that are not compliant if the cluster state permits it - strip common domain suffix from nodes and instances, so that output is shorter and hopefully clearer Version 0.0.4 (Sun, 15 Feb 2009) -------------------------------- - better balancing algorithm in hbal - implemented an RAPI collector, now the cluster data can be gathered automatically via RAPI and doesn't need manual export of node and instance list Version 0.0.3 (Wed, 28 Jan 2009) -------------------------------- - initial release of the hbal, a cluster rebalancing tool - input data format changed due to hbal requirements Version 0.0.2 (Tue, 06 Jan 2009) -------------------------------- - fix handling of some common cases (cluster N+1 compliant from the start, too big depth given, failure to compute solution) - add option to print the needed command list for reaching the proposed solution Version 0.0.1 (Tue, 06 Jan 2009) -------------------------------- - initial release of hn1 tool .. vim: set textwidth=72 : .. Local Variables: .. mode: rst .. fill-column: 72 .. End: