commit ee3986dc7034c863e4f76b1a25398708badafdba
author Primiano Tucci <primiano@google.com> Mon Aug 20 11:39:02 2018 +0200
committer Primiano Tucci <primiano@google.com> Mon Aug 20 11:39:02 2018 +0200
tree 7a2b452e43f759d068741e5ee9b804ef9eb118d1
parent c6b2cf82b09b883eb8fb142854bd85a7f1a3e021

TraceProcessor: Add magic ts_lower_bound to sched table

This CL introduces a new hidden column to the sched table
called ts_lower_bound. It allows to express lower-bound
searches on the table, without requiring full table scans.
The problem it solves is the following: imagine to have
4 slices in the storage and a visible time range in the UI
as follows:
  [       Slice A       ][ Slice B ][ Slice C ][ Slice D]
                   |------------------------------|
                   X      (Visible time range)    Y

The UI wants to see also the remainder of Slice A. In order to
do so, it cannot do a simple ts >= X query, because that will
miss Slice A. The right query would be
"WHERE ts + dur >= X AND ts <= y"
Unfortunately this query cannot use indices, due to the presence
of two columns (|ts| and |dur|) in the same constraint.
In order to achieve this, this CL introduces a |ts_lower_bound|
hidden column. When set, it has the following semantics: it filters
all slices with a timestamp greater or equal to:
- The greatest timestamp < |ts_lower_bound|, if it exists
- The smallest timestmap >= |ts_lower_bound| otherwise.
So the UI can instead query:
"WHERE ts_lower_bound = X and ts <= y"

Test: perfetto_unittests  --gtest_filter=SchedSliceTableTest.*
Change-Id: I3857fac37d713a126749194db965b84f537512dd