You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
38787: row: bugfixes to cfetcher r=jordanlewis a=jordanlewis
Fixes#38755.
Fixes#38611.
Fixes#38609.
Two bugfixes to the cfetcher.
- row: teach cfetcher about missing sentinel kvs
System tables are special in that they might be missing their "sentinel
kv": the kv that's marked as column family 0. Previously, cfetcher
assumed that a sentinel kv would always exist. This led to the inability
of decoding system tables correctly.
Now, if the table being read has more than one column family, we always
actively decode the column family id from the first key in a row to make
sure we know what values will be available in the kv.
- row: cfetcher bugfix to unique idxs w/ nulls
Previously, the cfetcher misbehaved when decoding unique secondary
indexes that had null values in one or more of their index columns. This
was because the encoding of the "extra columns" in a unique index (the
columns that are in the primary index but missing from the secondary) is
performed using the key encoding, not the value encoding, and the key
decoding routines in the cfetcher hadn't been taught to keep track of
which needed columns hadn't been seen in the current row yet.
This commit corrects the problem by teaching the key decoding routine to
optionally remove the decoded keys from the set of unseen column values
so far.
Release note: None
Co-authored-by: Jordan Lewis <[email protected]>
When
experimental_vectorize
is enabled at the cluster level, selecting fromcrdb_internal.zones
panics.The text was updated successfully, but these errors were encountered: