Elasticsearch Version

master

Installed Plugins

No response

Java Version

bundled

OS Version

all

Problem Description

TimeSeriesIndexSearcher use the tsidOrd to speed up tsid compare. There is a hidden bug:
The init value of tsidOrd is 0, so it's ok to collect data from beginning. But if collect data from the middle, the tsidOrd has a wrong init value. It will cause the PriorityQueue pop the wrong LeafWalker.

Steps to Reproduce

Create 2 segments IndexWriter, And the first segment is:

the second segment is:

the query is _tsid : tsid2 OR _tsid : tsid3
the current order is : tsid2/21 -> tsid2/11 -> tsid2/20 -> tsid2/10 -> tsid3/1 -> tsid3/1.
The right order is : tsid2/21 -> tsid2/20 -> tsid2/11 -> tsid2/10 -> tsid3/1 -> tsid3/1.
The reason is in

        boolean shouldPop() throws IOException {
            if (tsidOrd == -1) {
                tsidOrd = tsids.ordValue();
            } else if (tsidOrd != tsids.ordValue()) {
                tsidOrd = tsids.ordValue();
                return true;
            }
            return false;
        }

the tsidOrd init value is 0, But the first tsidOrd value of this query is 1. So It pop the second segment unexpected.

Logs (if relevant)

No response