On Wed, Jan 3, 2024 at 6:36 PM Nico Williams <n...@cryptonector.com> wrote: > On Tue, Jan 02, 2024 at 10:14:16AM -0500, Robert Haas wrote: > > It seems like a pretty significant savings no matter what. Suppose the > > backup_manifest file is 2GB, and instead of creating a 2GB buffer, you > > create an 1MB buffer and feed the data to the parser in 1MB chunks. > > Well, that saves 2GB less 1MB, full stop. Now if we address the issue > > you raise here in some way, we can potentially save even more memory, > > which is great, but even if we don't, we still saved a bunch of memory > > that could not have been saved in any other way. > > You could also build a streaming incremental parser. That is, one that > outputs a path and a leaf value (where leaf values are scalar values, > `null`, `true`, `false`, numbers, and strings). Then if the caller is > doing something JSONPath-like then the caller can probably immediately > free almost all allocations and even terminate the parse early.
I think our current parser is event-based rather than this ... but it seems like this could easily be built on top of it, if someone wanted to. -- Robert Haas EDB: http://www.enterprisedb.com