[HBASE-18294] Reduce global heap pressure: flush based on heap occupancy

• Type: Improvement
• Status: Resolved
• Priority: Major
• Resolution: Fixed
• Affects Version/s: 3.0.0
• Fix Version/s: 2.0.0-beta-2, 2.0.0

Release Note

A region is flushed if its memory component exceeds the region flush threshold.
A flush policy decides which stores to flush by comparing the size of the store to a column-family-flush threshold.
If the overall size of all memstores in the machine exceeds the bounds defined by the administrator (denoted global pressure) a region is selected and flushed.
HBASE-18294 changes flush decisions to be based on heap-occupancy and not data (key-value) size, consistently across levels. This rolls back some of the changes by HBASE-16747. Specifically,
(1) RSs, Regions and stores track their overall on-heap and off-heap occupancy,
(2) A region is flushed when its on-heap+off-heap size exceeds the region flush threshold specified in hbase.hregion.memstore.flush.size,
(3) The store to be flushed is chosen based on its on-heap+off-heap size
(4) At the RS level, a flush is triggered when the overall on-heap exceeds the on-heap limit, or when the overall off-heap size exceeds the off-heap limit (low/high water marks).

Note that when the region flush size is set to XXmb a region flush may be triggered even before writing keys and values of size XX because the total heap occupancy of the region which includes additional metadata exceeded the threshold.

• Assignee: Eshcar Hillel
• Reporter: Eshcar Hillel
• Created: 29/Jun/17 09:54
• Updated: 21/Mar/18 22:23
• Resolved: 19/Feb/18 13:51

Description

A region is flushed if its memory component exceed a threshold (default size is 128MB).
A flush policy decides whether to flush a store by comparing the size of the store to another threshold (that can be configured with hbase.hregion.percolumnfamilyflush.size.lower.bound).
Currently the implementation (in both cases) compares the data size (key-value only) to the threshold where it should compare the heap size (which includes index size, and metadata).

Issue Links

• links to: Glossary, concerns doc.
• links to: Review board