I think there's an order-minutes property which would cause a flush on idle.
table.compaction.minor.idle IIRC Jeff Kubina wrote:
Interesting, is it only flushed when the buffer is full or is there a time limit on it also? For example, if 25M of mutations are written and no more when is the buffer flushed? -- Jeff Kubina 410-988-4436 On Thu, Jul 7, 2016 at 11:50 AM, Christopher <[email protected] <mailto:[email protected]>> wrote: The change was introduced in https://issues.apache.org/jira/browse/ACCUMULO-1950, and it's an entirely new property. The old property was a per-session property. The new one is per-tserver, and is a better strategy, because it reduces the risk of multiple writers exhausting exhausting tserver memory, while still giving the user control over how frequently flushes/sync's occur. On Thu, Jul 7, 2016 at 10:32 AM Jeff Kubina <[email protected] <mailto:[email protected]>> wrote: I noticed that the default value for tserver.total.mutation.queue.max in 1.7 is 50M but in 1.6 it is 1M (tserver.mutation.queue.max). Is this increase to compensate for the performance hit of moving the WALs to the HDFS or some other factor? Is there a way to compute the number of times the buffer is flushed to calculate how this effects performance? -- Jeff Kubina
