summaryrefslogtreecommitdiffstats
path: root/notes
diff options
context:
space:
mode:
authorBryan Newbold <bnewbold@robocracy.org>2018-04-11 15:31:06 -0700
committerBryan Newbold <bnewbold@robocracy.org>2018-04-11 15:31:06 -0700
commit4b521b5eb0eb0843e365ac063cb04706d1cb674a (patch)
tree51a898579f1e1e11ce1261bcae205a41dab8840a /notes
parent229b22cedf786d55af210c806864459b29c1b27d (diff)
downloadfatcat-4b521b5eb0eb0843e365ac063cb04706d1cb674a.tar.gz
fatcat-4b521b5eb0eb0843e365ac063cb04706d1cb674a.zip
docs update
Diffstat (limited to 'notes')
-rw-r--r--notes/thoughts.txt32
1 files changed, 32 insertions, 0 deletions
diff --git a/notes/thoughts.txt b/notes/thoughts.txt
new file mode 100644
index 00000000..c01c0d37
--- /dev/null
+++ b/notes/thoughts.txt
@@ -0,0 +1,32 @@
+
+Instead of having a separate id pointer table, could have an extra "mutable"
+public ID column (unique, indexed) on entity rows. Backend would ensure the
+right thing happens. Changelog tables (or special redirect/deletion tables)
+would record changes and be "fallen through" to.
+
+Instead of having merge redirects, could just point all identifiers to the same
+revision (and update them all in the future). Don't need to recurse! Need to
+keep this forever though, could scale badly if "aggregations" get merged.
+
+Redirections of redirections should probably simply be disallowed.
+
+"Deletion" is really just pointing to a special or null entity.
+
+Trade-off: easy querying for common case (wanting "active" rows) vs. robust
+handling of redirects (likely to be pretty common). Also, having UUID handling
+across more than one table.
+
+## Scaling database
+
+Two scaling issues: size of database due to edits (likely billions of rows) and
+desire to do complex queries/reports ("analytics"). The later is probably not a
+concern, and could be handled by dumping and working on a cluster (or secondary
+views, etc). So just a distraction? Simpler to have all rolled up.
+
+Cockroach is postgres-like; might be able to use that for HA and scaling?
+Bottlenecks are probably complex joins (mitigated by "interleave"?) and bulk
+import performance (one-time?).
+
+Using elastic for most (eg, non-logged-in) views could keep things fast.
+
+Cockroach seems more resourced/polished than TiDB?