Closed knz closed 5 years ago
cc @mjibson thoughts? Is this a symptom of pgweb not supporting NaNs when it should, or is this a case of CockroachDB stuffing an incompatible value (NaN) in a type where clients don't expect it?
(In either case perhaps we can move the issue forward by replacing NaNs with NULLs for the internal table)
show create table crdb_internal.node_statement_statistics
indicates these columns are all floats, so NaN is a valid value here, suggesting this is a pgweb bug. Presumably we are returning NaN there because you can't have variance with only one value. I'm not convinced NULL is a better choice than NaN, from a math perspective.
I see thanks. I'll try a regular table with a nan in it and see what it says.
I have verified that pgweb indeed fails with a 500 error on regular tables containing a float column with a NaN value. So this is non-actionable.
Note: this may not be a problem inside CockroachDB and instead a bug in pgweb. Unsure.
However since pgweb is rather consistently good at viewing CockroachDB we should do moderate effort at ensuring the experience remains smooth.
Describe the problem
When (trying to) inspecting
crdb_internal.node_statement_statistics
the inspection fails and pgweb returns an internal error 500 instead, with the following stack trace:To Reproduce
(nothing happens in the browser)
Expected behavior
either a successful visualization or a better error message.