[PGPRO-5691] move mmapped ptrack map into shared postgres memory

Author: kulaginm

README.md

@@ -57,7 +57,7 @@ postgres=# CREATE EXTENSION ptrack;
 
 ## Configuration
 
-The only one configurable option is `ptrack.map_size` (in MB). Default is `-1`, which means `ptrack` is turned off. In order to reduce number of false positives it is recommended to set `ptrack.map_size` to `1 / 1000` of expected `PGDATA` size (i.e. `1000` for a 1 TB database).
+The only one configurable option is `ptrack.map_size` (in MB). Default is `0`, which means `ptrack` is turned off. In order to reduce number of false positives it is recommended to set `ptrack.map_size` to `1 / 1000` of expected `PGDATA` size (i.e. `1000` for a 1 TB database).
 
 To disable `ptrack` and clean up all remaining service files set `ptrack.map_size` to `0`.
 
@@ -74,7 +74,7 @@ Usage example:
 postgres=# SELECT ptrack_version();
  ptrack_version 
 ----------------
- 2.2
+ 2.3
 (1 row)
 
 postgres=# SELECT ptrack_init_lsn();
@@ -115,27 +115,29 @@ Usually, you have to only install new version of `ptrack` and do `ALTER EXTENSIO
 
 Since version 2.2 we use a different algorithm for tracking changed pages. Thus, data recorded in the `ptrack.map` using pre 2.2 versions of `ptrack` is incompatible with newer versions. After extension upgrade and server restart old `ptrack.map` will be discarded with `WARNING` and initialized from the scratch.
 
+#### Upgrading from 2.2.* to 2.3.*:
+TODO
+
 ## Limitations
 
 1. You can only use `ptrack` safely with `wal_level >= 'replica'`. Otherwise, you can lose tracking of some changes if crash-recovery occurs, since [certain commands are designed not to write WAL at all if wal_level is minimal](https://www.postgresql.org/docs/12/populate.html#POPULATE-PITR), but we only durably flush `ptrack` map at checkpoint time.
 
 2. The only one production-ready backup utility, that fully supports `ptrack` is [pg_probackup](https://github.com/postgrespro/pg_probackup).
 
-3. Currently, you cannot resize `ptrack` map in runtime, only on postmaster start. Also, you will loose all tracked changes, so it is recommended to do so in the maintainance window and accompany this operation with full backup. See [TODO](#TODO) for details.
+3. You cannot resize `ptrack` map in runtime, only on postmaster start. Also, you will loose all tracked changes, so it is recommended to do so in the maintainance window and accompany this operation with full backup.
 
-4. You will need up to `ptrack.map_size * 3` of additional disk space, since `ptrack` uses two additional temporary files for durability purpose. See [Architecture section](#Architecture) for details.
+4. You will need up to `ptrack.map_size * 2` of additional disk space, since `ptrack` uses additional temporary file for durability purpose. See [Architecture section](#Architecture) for details.
 
 ## Benchmarks
 
 Briefly, an overhead of using `ptrack` on TPS usually does not exceed a couple of percent (~1-3%) for a database of dozens to hundreds of gigabytes in size, while the backup time scales down linearly with backup size with a coefficient ~1. It means that an incremental `ptrack` backup of a database with only 20% of changed pages will be 5 times faster than a full backup. More details [here](benchmarks).
 
 ## Architecture
 
-We use a single shared hash table in `ptrack`, which is mapped in memory from the file on disk using `mmap`. Due to the fixed size of the map there may be false positives (when some block is marked as changed without being actually modified), but not false negative results. However, these false postives may be completely eliminated by setting a high enough `ptrack.map_size`.
+We use a single shared hash table in `ptrack`. Due to the fixed size of the map there may be false positives (when some block is marked as changed without being actually modified), but not false negative results. However, these false postives may be completely eliminated by setting a high enough `ptrack.map_size`.
 
-All reads/writes are made using atomic operations on `uint64` entries, so the map is completely lockless during the normal PostgreSQL operation. Because we do not use locks for read/write access and cannot control `mmap` eviction back to disk, `ptrack` keeps a map (`ptrack.map`) since the last checkpoint intact and uses up to 2 additional temporary files:
+All reads/writes are made using atomic operations on `uint64` entries, so the map is completely lockless during the normal PostgreSQL operation. Because we do not use locks for read/write access, `ptrack` keeps a map (`ptrack.map`) since the last checkpoint intact and uses up to 1 additional temporary file:
 
-* working copy `ptrack.map.mmap` for doing `mmap` on it (there is a [TODO](#TODO) item);
 * temporary file `ptrack.map.tmp` to durably replace `ptrack.map` during checkpoint.
 
 Map is written on disk at the end of checkpoint atomically block by block involving the CRC32 checksum calculation that is checked on the next whole map re-read after crash-recovery or restart.
@@ -165,8 +167,6 @@ Available test modes (`MODE`) are `basic` (default) and `paranoia` (per-block ch
 
 ### TODO
 
-* Use POSIX `shm_open()` instead of `open()` to do not create an additional working copy of `ptrack` map file.
 * Should we introduce `ptrack.map_path` to allow `ptrack` service files storage outside of `PGDATA`? Doing that we will avoid patching PostgreSQL binary utilities to ignore `ptrack.map.*` files.
 * Can we resize `ptrack` map on restart but keep the previously tracked changes?
-* Can we resize `ptrack` map dynamicaly?
 * Can we write a formal proof, that we never loose any modified page with `ptrack`? With TLA+?