PiTR: cannot restore index created during log backup if it is longer than 3072

[YuJuncen]

Bug Report

Please answer these questions before submitting your issue. Thanks!

1. Minimal reproduce step (Required)

Start a cluster with:

tiup playground nightly --tiflash=0 --db.config=./op/adhoc/index-length-8001.toml

Where the config index-length-8001.toml is:

max-index-length = 8001

Do the following:

mycli --port 4000 --execute "DROP TABLE IF EXISTS test.huge_idx;"
mycli --port 4000 --execute "CREATE TABLE test.huge_idx(id int AUTO_INCREMENT, blob1 varchar(1000), blob2 varchar(1000))"

$br backup full -s (s3path "$strg_prefix-full") $argv
set backupts (mcli cat "loc/breeze/$strg_prefix-full/backupmeta" | _get_brpb BackupMeta | rg -r '$1' 'end_version: (\d+)$')
$br log start -s (s3path "$strg_prefix-incr") --task-name test $argv --start-ts "$backupts"

mycli --port 4000 --execute "CREATE INDEX huge ON test.huge_idx(blob1, blob2);"

# Wait until the checkpoint advances...
$br restore point -s (s3path huge_index-incr) --full-backup-storage (s3path huge_index-full)

2. What did you expect to see? (Required)

Restore should success.

3. What did you see instead (Required)

repair ingest index huge for table test.huge_idx  ABORTED
["restore log failed summary"] [error="failed to repair ingest index: [ddl:1071]Specified key was too long (4000 bytes); max key length is 3072 bytes"]

4. What is your TiDB version? (Required)

nightly