The map::at method does not always exist, so instead use map::find.

[cumulus.git] / schema.sql

diff --git a/schema.sql b/schema.sql
index a37b501..35b2c9d 100644 (file)
--- a/schema.sql
+++ b/schema.sql
@@ -7,8 +7,9 @@
 create table snapshots (
     snapshotid integer primary key,
     name text not null,
-    scheme text,
-    timestamp real
+    scheme text not null,
+    timestamp real,
+    intent real
 );
 
 -- List of segments which have been created.
@@ -16,10 +17,13 @@ create table segments (
     segmentid integer primary key,
     segment text unique not null,
     path text,
-    checksum text
+    checksum text,
+    mtime real,
+    size integer,
+    expire_time integer         -- snapshotid of latest snapshot when expired
 );
 
--- Index of all blocks which have been stored in a snapshot, by checksum.
+-- Index of all blocks which have been stored, by checksum.
 create table block_index (
     blockid integer primary key,
     segmentid integer not null,
@@ -32,28 +36,41 @@ create table block_index (
 create index block_content_index on block_index(checksum);
 create unique index block_name_index on block_index(segmentid, object);
 
--- Index tracking which blocks are used by which snapshots.
-create table snapshot_contents (
-    blockid integer,
-    snapshotid integer
+-- Checksums for the decomposition of blocks into even smaller chunks
+-- (variable-sized, but generally ~4 kB, and maximum 64 kB).  Chunk boundaries
+-- are determined based on the contents using Rabin fingerprints.  These
+-- checksums can be used for computing sub-file incrementals.
+--
+-- Each block stored in block_index may have an entry in the
+-- subblock_signatures table.  The signatures field is a binary blob consisting
+-- of a packed sequence of (chunk length [16-bit unsigned, big-endian],
+-- checksum [20 bytes if SHA-1]) tuples that should cover the entire block.
+--
+-- algorithm specifies the method used for computing break points as well as
+-- the hash function used, so that signatures can be discarded if the algorithm
+-- changes.  The current algorithm used is 'lbfs-4096/sha1', which specifies a
+-- target 4 kB block size with parameters set to match LBFS, and SHA-1 as the
+-- hash algorithm.
+create table subblock_signatures (
+    blockid integer primary key,
+    algorithm text not null,
+    signatures blob not null
 );
-create unique index snapshot_contents_unique
-    on snapshot_contents(blockid, snapshotid);
 
--- Summary statistics for each segment.
-create view segment_info as select * from
-    (select segmentid, max(timestamp) as mtime,
-            sum(size) as size, count(*) as objects
-       from block_index natural join segments group by segmentid)
-natural join
-    (select segmentid, sum(size) as used, count(*) as objects_used
-       from block_index where blockid in
-            (select blockid from snapshot_contents) group by segmentid);
+-- Summary of segment utilization for each snapshots.
+create table segments_used (
+    snapshotid integer not null,
+    segmentid integer not null,
+    utilization real
+);
+create unique index segments_used_index
+    on segments_used(snapshotid, segmentid);
 
--- Ranking of segments to be cleaned, using a benefit function of
--- (fraction free space)*(age of youngest object).
-create view cleaning_order as select *, (1-u)*age/(u+0.1) as benefit from
-    (select segmentid,
-            cast(used as real) / size as u, julianday('now') - mtime as age
-        from segment_info)
-where benefit > 0;
+-- Overall estimate of segment utilization, for all snapshots combined.
+create view segment_info as
+select segmentid, mtime, size, expire_time,
+       cast(size * utilization as integer) as used, utilization
+from segments join
+     (select segmentid, max(utilization) as utilization
+      from segments_used group by segmentid)
+using (segmentid);