Merge pull request #1500 from joto/node-location-for-pgsql-middle

Use new node_location_t in pgsql middle as cache

Author: lonvia

docs/osm2pgsql.1

@@ -202,25 +202,7 @@ Defaults to 800.
 .RE
 .TP
 .B \-\-cache\-strategy=STRATEGY
-There are a number of different modes in which osm2pgsql can organize
-its node cache in RAM.
-These are optimized for different assumptions of the data and the
-hardware resources available.
-Currently available strategies are \f[B]dense\f[], \f[B]chunked\f[],
-\f[B]sparse\f[] and \f[B]optimized\f[].
-\f[B]dense\f[] assumes that the node id numbers are densely packed,
-i.e.\ only a few IDs in the range are missing / deleted.
-For planet extracts this is usually not the case, making the cache very
-inefficient and wasteful of RAM.
-\f[B]sparse\f[] assumes node IDs in the data are not densely packed,
-greatly increasing caching efficiency in these cases.
-If node IDs are densely packed, like in the full planet, this strategy
-has a higher overhead for indexing the cache.
-\f[B]optimized\f[] uses both dense and sparse strategies for different
-ranges of the ID space.
-On a block by block basis it tries to determine if it is more effective
-to store the block of IDs in sparse or dense mode.
-This is the default and should be typically used.
+This deprecated option will be ignored.
 .RS
 .RE
 .TP

docs/osm2pgsql.md

@@ -144,19 +144,7 @@ mandatory for short options too.
     Defaults to 800.
 
 \--cache-strategy=STRATEGY
-:   There are a number of different modes in which osm2pgsql can organize its
-    node cache in RAM. These are optimized for different assumptions of the data
-    and the hardware resources available. Currently available strategies are
-    **dense**, **chunked**, **sparse** and **optimized**. **dense** assumes
-    that the node id numbers are densely packed, i.e. only a few IDs in the range are
-    missing / deleted. For planet extracts this is usually not the case, making the cache
-    very inefficient and wasteful of RAM. **sparse** assumes node IDs in the data
-    are not densely packed, greatly increasing caching efficiency in these cases.
-    If node IDs are densely packed, like in the full planet, this strategy has a higher
-    overhead for indexing the cache. **optimized** uses both dense and sparse strategies
-    for different ranges of the ID space. On a block by block basis it tries to determine
-    if it is more effective to store the block of IDs in sparse or dense mode. This is the
-    default and should be typically used.
+:   This deprecated option will be ignored.
 
 -x, \--extra-attributes
 :   Include attributes of each object in the middle tables and make them

src/CMakeLists.txt

@@ -13,7 +13,6 @@ set(osm2pgsql_lib_SOURCES
   middle-ram.cpp
   node-locations.cpp
   node-persistent-cache.cpp
-  node-ram-cache.cpp
   options.cpp
   ordered-index.cpp
   osmdata.cpp

src/middle-pgsql.cpp

@@ -32,7 +32,6 @@
 #include "logging.hpp"
 #include "middle-pgsql.hpp"
 #include "node-persistent-cache.hpp"
-#include "node-ram-cache.hpp"
 #include "options.hpp"
 #include "osmtypes.hpp"
 #include "pgsql-helper.hpp"
@@ -587,7 +586,7 @@ void middle_pgsql_t::after_relations()
 }
 
 middle_query_pgsql_t::middle_query_pgsql_t(
-    std::string const &conninfo, std::shared_ptr<node_ram_cache> const &cache,
+    std::string const &conninfo, std::shared_ptr<node_locations_t> const &cache,
     std::shared_ptr<node_persistent_cache> const &persistent_cache)
 : m_sql_conn(conninfo), m_cache(cache), m_persistent_cache(persistent_cache)
 {
@@ -782,8 +781,8 @@ static bool check_bucket_index(pg_conn_t *db_connection,
 
 middle_pgsql_t::middle_pgsql_t(options_t const *options)
 : m_options(options),
-  m_cache(new node_ram_cache{options->alloc_chunkwise | ALLOC_LOSSY,
-                             options->cache}),
+  m_cache(new node_locations_t{static_cast<std::size_t>(options->cache) *
+                               1024UL * 1024UL}),
   m_db_connection(m_options->database_options.conninfo()),
   m_copy_thread(
       std::make_shared<db_copy_thread_t>(options->database_options.conninfo())),

src/middle-pgsql.hpp

@@ -24,8 +24,8 @@
 
 #include "db-copy-mgr.hpp"
 #include "middle.hpp"
+#include "node-locations.hpp"
 #include "node-persistent-cache.hpp"
-#include "node-ram-cache.hpp"
 #include "options.hpp"
 #include "pgsql.hpp"
 
@@ -34,7 +34,7 @@ class middle_query_pgsql_t : public middle_query_t
 public:
     middle_query_pgsql_t(
         std::string const &conninfo,
-        std::shared_ptr<node_ram_cache> const &cache,
+        std::shared_ptr<node_locations_t> const &cache,
         std::shared_ptr<node_persistent_cache> const &persistent_cache);
 
     size_t nodes_get_list(osmium::WayNodeList *nodes) const override;
@@ -55,7 +55,7 @@ class middle_query_pgsql_t : public middle_query_t
     std::size_t get_way_node_locations_db(osmium::WayNodeList *nodes) const;
 
     pg_conn_t m_sql_conn;
-    std::shared_ptr<node_ram_cache> m_cache;
+    std::shared_ptr<node_locations_t> m_cache;
     std::shared_ptr<node_persistent_cache> m_persistent_cache;
 };
 
@@ -147,7 +147,7 @@ struct middle_pgsql_t : public middle_t
 
     options_t const *m_options;
 
-    std::shared_ptr<node_ram_cache> m_cache;
+    std::shared_ptr<node_locations_t> m_cache;
     std::shared_ptr<node_persistent_cache> m_persistent_cache;
 
     pg_conn_t m_db_connection;