Hi Donald, This is a destruction ordering problem for static variables. Unfortunately we don't have control over all the static variables, so we can't guaranteed the order of destruction. For example, the logging system that we use (log4cpp) has a static variable that it depends on. If the logging system shuts down before some other object gets destructed and part of the destruction sequence involves printing a log message, the process will dump core. We plan to look into the possibility of eliminating all static variables in Hypertable, but until then, we're terminating all programs with _exit, which eliminates spurious core dumps on program exit.
- Doug On Thu, Mar 5, 2009 at 7:57 AM, Liu Kejia (Donald) <[email protected]>wrote: > Sorry for the empty message.. > > * Changed a number of programs to exit using _exit to avoid static > destruction problems > > > Is this because of destructors that throw exceptions? I think we should > avoid "throwing destructors" because it is bad-bad C++ practice, instead of > avoiding it using _exit > > > On Thu, Mar 5, 2009 at 11:53 PM, Liu Kejia (Donald) > <[email protected]>wrote: > >> >> >> On Thu, Mar 5, 2009 at 5:20 AM, Doug Judd <[email protected]> wrote: >> >>> Patch release 0.9.2.2 has been released and is now available for >>> download at http://www.hypertable.org/download.html. >>> >>> NOTE: you may have to hit refresh in your browser to get the correct >>> download >>> >>> Version 0.9.2.2: >>> (2009-03-04) >>> >>> * Added support for ALTER TABLE >>> * Addded Bloom Filter to CellStore >>> * Added calls to _exit in random_read/write benchmark programs to avoid >>> static destruction >>> * [issue 251] Eliminated race condition between load_range and >>> drop_table >>> * Upgraded to Hadoop 0.19.1 >>> * [issue 245] Fixed segfault problem with splits and multiple AGs >>> * Updated thrift support for new namespace changes in the upstream >>> library >>> * Added missing columns in the thrift scan spec >>> * Added more scanner/mutator examples for various languages >>> * Updated README link to the new recommended thrift snapshot >>> * Made CellStore::create API more generic >>> * Added row_len and column_qualifier_len in Key to save repeated >>> strlens >>> * Enabled Bloom filter by default >>> * Fixed crash in TableMutator if global config properties are not >>> initialized >>> * Added floating point support for properties >>> * Fixed broken each_cell and with_mutator methods in ruby thrift client >>> * Wrapped with_* methods in ensure block >>> * Added tests for scanner and mutator usage >>> * Issue 229: fixed some warnings with gcc 4.3+ >>> * [issue 239] Made serverup respect --wait value in all places >>> * [issue 246] Fixed commit log GC problem >>> * Upgraded to Quicklz 1.4 and made quicklz the default compressor for >>> CommitLog >>> * [issue 245] Use broker-assigned strictly increasing fds in >>> KosmosBroker >>> * [issue 234] Made wait for Hyperspace respect supplied timeout >>> * [issue 127] Made Hyperspace server detect suspension and extend >>> leases >>> * [issue 231] Fixed problem where CellStoreV0::add return value was >>> getting ignored >>> * [issue 238] Fixed crash problem during RangeServer::replay_update >>> * [issue 235] Eliminated abort after split attempts fail >>> * [issue 236] Fixed hang problem on CentOS 5 >>> * Fixed a range server restart issue by overwriting tmp explicitly >>> * Added wait logic in Master when root METADATA range server is down >>> * Fixed problem where drop_table was executed during a range split >>> causing bad RSML sequence >>> * Got rid of an assert in the Comm layer on bad header >>> * Fixed problem where 'Files' column not updated properly after split >>> * Fixed problem where Gc.Interval was being treated as seconds >>> * Changed a number of programs to exit using _exit to avoid static >>> destruction problems >>> * Fixed problem caused by invalid initializations of rssstat (Hiroyuki) >>> * Made stopping servers faster by waiting in parallel >>> * Leave RSML alone in case of errors >>> * Refactored bash specific commands in Capfile into a separate script >>> * Disable thrift broker API logging by default >>> * Issue 190: Reverted to the METADATA sanity check since RangeLocator >>> is fixed >>> >>> >>> >>> >> > > > > --~--~---------~--~----~------------~-------~--~----~ You received this message because you are subscribed to the Google Groups "Hypertable Development" group. To post to this group, send email to [email protected] To unsubscribe from this group, send email to [email protected] For more options, visit this group at http://groups.google.com/group/hypertable-dev?hl=en -~----------~----~----~----~------~----~------~--~---
