Hi Dorian, Thanks! The rationale for the SplitSize change is that the original split size came from the Bigtable paper which was written in 2006. The capacity of commodity hardware has increased significantly since then, so doubling the Range size seems appropriate. Many of our deployments set the Split size at 1GB or 2GB. It's best to cap the Range size, otherwise the entire system can get bogged down as it compacts big ranges. We've thought a lot about whether or not we can do away with this limitation and split rows across ranges, but have decided that it's not really feasible from a query execution perspective. The limitation of requiring a row to fit on a single Range is definitely a weakness which we will address at some point in the near future.
- Doug On Fri, Aug 16, 2013 at 5:39 AM, ddorian <[email protected]> wrote: > CONTGRATS! > > Curious why you changed the range_size (because of big rows or big ranges, > or both). > > Also is it better to let rows grow than to just not accept writes since at > least it will not break applications? > > > On Friday, August 16, 2013 7:30:07 AM UTC+2, Doug Judd wrote: >> >> Hypertable version 0.9.7.9 is now available for download at >> http://www.hypertable.com/**download/<http://www.hypertable.com/download/> >> . >> >> Version 0.9.7.9: >> (2013-08-15) >> >> QFS Broker >> Improved disk usage monitoring and handling >> Added system state with READONLY variable and HQL SET command >> Changed Hypertable.RangeServer.Range.**SplitSize default to 512MB >> Fixed bug in delete processing in MergeScannerAccessGroup >> Fixed CELL_LIMIT processing >> Fixed level calculation in MaintenanceQueue >> Reduced calls to Range::get_maintenance_data() in get_statistics() >> Added version, start_row, and end_row to hints file >> Moved transfer log to corresponding /tables/ directory >> Fixed race condition in creation of RemoveOkLogs entity >> Changed "dropping task" log message from WARN to INFO >> Fixed race condition in RangeServer-failover-restart test >> Better error state capture for group commit and rs failover basic tests >> Replaced RSStats with LoadStatistics class >> Added server disk usage statistics to RS_METRICS >> Fix tests >> Added PosixReaddir to DfsBroker interface >> >> >> -- >> Doug Judd >> CEO, Hypertable Inc. >> > -- > You received this message because you are subscribed to the Google Groups > "Hypertable Development" group. > To unsubscribe from this group and stop receiving emails from it, send an > email to [email protected]. > To post to this group, send email to [email protected]. > Visit this group at http://groups.google.com/group/hypertable-dev. > For more options, visit https://groups.google.com/groups/opt_out. > -- Doug Judd CEO, Hypertable Inc. -- You received this message because you are subscribed to the Google Groups "Hypertable Development" group. To unsubscribe from this group and stop receiving emails from it, send an email to [email protected]. To post to this group, send email to [email protected]. Visit this group at http://groups.google.com/group/hypertable-dev. For more options, visit https://groups.google.com/groups/opt_out.
