Merge pull request #14 from marselester/lookup-cache

Lookup cache

Author: marselester

.github/workflows/ci.yml

@@ -25,7 +25,7 @@ jobs:
         run: zig build test
       - name: Run lookup example
         run: zig build example_lookup
-      - name: Run within example
-        run: zig build example_within
+      - name: Run scan example
+        run: zig build example_scan
       - name: Run inspect example
         run: zig build example_inspect

README.md

@@ -54,13 +54,37 @@ var db = try maxminddb.Reader.mmap(allocator, db_path, .{ .ipv4_index_first_n_bi
 defer db.close();
 ```
 
-Use `ArenaAllocator` for best performance, see [benchmarks](./benchmarks/).
+Each `lookup` result owns an arena with all decoded allocations.
+Call `deinit()` to free it or use `ArenaAllocator` with `reset()`,
+see [benchmarks](./benchmarks/lookup.zig).
+
+```zig
+if (try db.lookup(maxminddb.geolite2.City, allocator, ip, .{})) |result| {
+    defer result.deinit();
+    std.debug.print("{f} {s}\n", .{ result.network, result.value.city.names.?.get("en").? });
+}
+
+var arena = std.heap.ArenaAllocator.init(allocator);
+defer arena.deinit();
+
+const arena_allocator = arena.allocator();
+for (ips) |ip| {
+    if (try db.lookup(maxminddb.geolite2.City, arena_allocator, ip, .{})) |result| {
+        std.debug.print("{f} {s}\n", .{ result.network, result.value.city.names.?.get("en").? });
+    }
+
+    _ = arena.reset(.retain_capacity);
+}
+```
 
 If you don't need all the fields, use `.only` to decode only the top-level fields you want.
 
 ```zig
 const fields = &.{ "city", "country" };
-const city = try db.lookup(allocator, maxminddb.geolite2.City, ip, .{ .only = fields });
+if (try db.lookup(maxminddb.geolite2.City, allocator, ip, .{ .only = fields })) |result| {
+    defer result.deinit();
+    std.debug.print("{f} {s}\n", .{ result.network, result.value.city.names.?.get("en").? });
+}
 ```
 
 Alternatively, define your own struct with only the fields you need.
@@ -74,27 +98,42 @@ const MyCity = struct {
     } = .{},
 };
 
-const city = try db.lookup(allocator, MyCity, ip, .{});
+if (try db.lookup(MyCity, allocator, ip, .{})) |result| {
+    defer result.deinit();
+    std.debug.print("{s}\n", .{result.value.city.names.en});
+}
 ```
 
 Use `any.Value` to decode any record without knowing the schema.
 
 ```zig
-const result = try db.lookup(allocator, maxminddb.any.Value, ip, .{ .only = fields });
-if (result) |r| {
+if (try db.lookup(maxminddb.any.Value, allocator, ip, .{ .only = fields })) |result| {
+    defer result.deinit();
     // Formats as compact JSON.
-    std.debug.print("{f}\n", .{r.value});
+    std.debug.print("{f}\n", .{result.value});
+}
+```
+
+Use `lookupWithCache` to skip decoding when different IPs resolve to the same record.
+The cache owns decoded memory, so results don't need to be individually freed.
+
+```zig
+var cache = try maxminddb.Cache(maxminddb.geolite2.City).init(allocator, .{});
+defer cache.deinit();
+
+if (try db.lookupWithCache(maxminddb.geolite2.City, &cache, ip, .{})) |result| {
+    std.debug.print("{f} {s}\n", .{ result.network, result.value.city.names.?.get("en").? });
 }
 ```
 
 Here are reference results on Apple M2 Pro (1M random IPv4 lookups against GeoLite2-City
 with `ipv4_index_first_n_bits = 16`):
 
-| Benchmark       | All fields | Filtered (city) |
-|---              |---         |---              |
-| `geolite2.City` | ~1,284,000 | ~1,348,000      |
-| `MyCity`        | ~1,383,000 | —               |
-| `any.Value`     | ~1,254,000 | ~1,349,000      |
+| Type            | Default    | `.only`    | `Cache`    |
+|---              |---         |---         |---         |
+| `geolite2.City` | ~1,420,000 | ~1,348,000 | ~1,565,000 |
+| `MyCity`        | ~1,383,000 |            |            |
+| `any.Value`     | ~1,254,000 | ~1,349,000 |            |
 
 <details>
 
@@ -140,6 +179,30 @@ Lookups Per Second (avg):1315870.3443053183
 
 <details>
 
+<summary>geolite2.City with Cache</summary>
+
+```sh
+$ for i in $(seq 1 10); do
+    zig build benchmark_lookup_cache -Doptimize=ReleaseFast -- GeoLite2-City.mmdb 1000000 \
+      2>&1 | grep 'Lookups Per Second'
+  done
+
+Lookups Per Second (avg):1493822.3908664712
+Lookups Per Second (avg):1503051.0049070602
+Lookups Per Second (avg):1499514.437731375
+Lookups Per Second (avg):1491749.9700251492
+Lookups Per Second (avg):1449924.9391983037
+Lookups Per Second (avg):1396100.6211600688
+Lookups Per Second (avg):1465750.9875955326
+Lookups Per Second (avg):1515611.9396877384
+Lookups Per Second (avg):1485235.6423035355
+Lookups Per Second (avg):1439334.222943596
+```
+
+</details>
+
+<details>
+
 <summary>MyCity</summary>
 
 ```sh
@@ -203,3 +266,83 @@ Lookups Per Second (avg):1315986.2950186788
 ```
 
 </details>
+
+Use `scan` to iterate over all networks in the database.
+
+```zig
+var it = try db.scan(maxminddb.any.Value, allocator, maxminddb.Network.all_ipv6, .{});
+
+while (try it.next()) |item| {
+    defer item.deinit();
+    std.debug.print("{f} {f}\n", .{ item.network, item.value });
+}
+```
+
+Use `scanWithCache` to avoid re-decoding networks that share the same record.
+The cache owns decoded memory, so results don't need to be individually freed.
+
+```zig
+var cache = try maxminddb.Cache(maxminddb.any.Value).init(allocator, .{});
+defer cache.deinit();
+
+var it = try db.scanWithCache(maxminddb.any.Value, &cache, maxminddb.Network.all_ipv6, .{});
+
+while (try it.next()) |item| {
+    std.debug.print("{f} {f}\n", .{ item.network, item.value });
+}
+```
+
+Here are reference results on Apple M2 Pro (full GeoLite2-City scan using `any.Value`):
+
+| Mode    | Records/sec |
+|---      |---          |
+| Default | ~1,295,000  |
+| `Cache` | ~2,930,000  |
+
+<details>
+
+<summary>no cache (any.Value)</summary>
+
+```sh
+$ for i in $(seq 1 10); do
+    zig build benchmark_scan -Doptimize=ReleaseFast -- GeoLite2-City.mmdb \
+      2>&1 | grep 'Records Per Second'
+  done
+
+Records Per Second: 1216758.945145436
+Records Per Second: 1238440.9772222256
+Records Per Second: 1234710.6362391203
+Records Per Second: 1229527.4688849829
+Records Per Second: 1243478.3908140333
+Records Per Second: 1226863.3718734735
+Records Per Second: 1240073.3248202254
+Records Per Second: 1247541.1528026997
+Records Per Second: 1230510.441029532
+Records Per Second: 1246311.587919839
+```
+
+</details>
+
+<details>
+
+<summary>cache (any.Value)</summary>
+
+```sh
+$ for i in $(seq 1 10); do
+    zig build benchmark_scan_cache -Doptimize=ReleaseFast -- GeoLite2-City.mmdb \
+      2>&1 | grep 'Records Per Second'
+  done
+
+Records Per Second: 2847560.3756875996
+Records Per Second: 2925388.867798729
+Records Per Second: 2919203.9046571665
+Records Per Second: 2814410.555872645
+Records Per Second: 2933972.04386147
+Records Per Second: 2900700.06160036
+Records Per Second: 2922279.338699886
+Records Per Second: 2862525.847598088
+Records Per Second: 2916760.542913819
+Records Per Second: 2908245.98918392
+```
+
+</details>

benchmarks/inspect.zig

@@ -59,8 +59,8 @@ pub fn main() !void {
         const ip = std.net.Address.initIp4(ip_bytes, 0);
 
         const result = db.lookup(
-            arena_allocator,
             maxminddb.any.Value,
+            arena_allocator,
             ip,
             .{ .only = fields },
         ) catch |err| {

benchmarks/lookup.zig

@@ -59,8 +59,8 @@ pub fn main() !void {
         const ip = std.net.Address.initIp4(ip_bytes, 0);
 
         const result = db.lookup(
-            arena_allocator,
             maxminddb.geolite2.City,
+            arena_allocator,
             ip,
             .{ .only = fields },
         ) catch |err| {

benchmarks/lookup_cache.zig

@@ -0,0 +1,91 @@
+const std = @import("std");
+const maxminddb = @import("maxminddb");
+
+const default_db_path: []const u8 = "GeoLite2-City.mmdb";
+const default_num_lookups: u64 = 1_000_000;
+const max_mmdb_fields = 32;
+
+pub fn main() !void {
+    const allocator = std.heap.smp_allocator;
+
+    const args = try std.process.argsAlloc(allocator);
+    defer std.process.argsFree(allocator, args);
+
+    var db_path: []const u8 = default_db_path;
+    var num_lookups = default_num_lookups;
+    var fields: ?[]const []const u8 = null;
+    if (args.len > 1) db_path = args[1];
+    if (args.len > 2) num_lookups = try std.fmt.parseUnsigned(u64, args[2], 10);
+    if (args.len > 3) {
+        var items: [max_mmdb_fields][]const u8 = undefined;
+
+        var it = std.mem.splitScalar(u8, args[3], ',');
+        var i: usize = 0;
+        while (it.next()) |part| : (i += 1) {
+            items[i] = part;
+        }
+
+        fields = items[0..i];
+    }
+
+    std.debug.print("Benchmarking with:\n", .{});
+    std.debug.print("  Database: {s}\n", .{db_path});
+    std.debug.print("  Lookups:  {d}\n", .{num_lookups});
+    std.debug.print("Opening database...\n", .{});
+
+    var open_timer = try std.time.Timer.start();
+    var db = try maxminddb.Reader.mmap(allocator, db_path, .{ .ipv4_index_first_n_bits = 16 });
+    defer db.close();
+    const open_time_ms = @as(f64, @floatFromInt(open_timer.read())) /
+        @as(f64, @floatFromInt(std.time.ns_per_ms));
+    std.debug.print("Database opened successfully in {d} ms. Type: {s}\n", .{
+        open_time_ms,
+        db.metadata.database_type,
+    });
+
+    var cache = try maxminddb.Cache(maxminddb.geolite2.City).init(allocator, .{});
+    defer cache.deinit();
+
+    std.debug.print("Starting benchmark...\n", .{});
+    var timer = try std.time.Timer.start();
+    var not_found_count: u64 = 0;
+    var lookup_errors: u64 = 0;
+    var ip_bytes: [4]u8 = undefined;
+
+    for (0..num_lookups) |_| {
+        std.crypto.random.bytes(&ip_bytes);
+        const ip = std.net.Address.initIp4(ip_bytes, 0);
+
+        const result = db.lookupWithCache(
+            maxminddb.geolite2.City,
+            &cache,
+            ip,
+            .{ .only = fields },
+        ) catch |err| {
+            std.debug.print("! Lookup error for IP {any}: {any}\n", .{ ip, err });
+            lookup_errors += 1;
+            continue;
+        };
+        if (result == null) {
+            not_found_count += 1;
+            continue;
+        }
+    }
+
+    const elapsed_ns = timer.read();
+    const elapsed_s = @as(f64, @floatFromInt(elapsed_ns)) /
+        @as(f64, @floatFromInt(std.time.ns_per_s));
+    const lookups_per_second = if (elapsed_s > 0)
+        @as(f64, @floatFromInt(num_lookups)) / elapsed_s
+    else
+        0.0;
+    const successful_lookups = num_lookups - not_found_count - lookup_errors;
+
+    std.debug.print("\n--- Benchmark Finished ---\n", .{});
+    std.debug.print("Total Lookups Attempted: {d}\n", .{num_lookups});
+    std.debug.print("Successful Lookups:      {d}\n", .{successful_lookups});
+    std.debug.print("IPs Not Found:           {d}\n", .{not_found_count});
+    std.debug.print("Lookup Errors:           {d}\n", .{lookup_errors});
+    std.debug.print("Elapsed Time:            {d} s\n", .{elapsed_s});
+    std.debug.print("Lookups Per Second (avg):{d}\n", .{lookups_per_second});
+}

benchmarks/mycity.zig

@@ -53,8 +53,8 @@ pub fn main() !void {
         const ip = std.net.Address.initIp4(ip_bytes, 0);
 
         const result = db.lookup(
-            arena_allocator,
             MyCity,
+            arena_allocator,
             ip,
             .{},
         ) catch |err| {

benchmarks/scan.zig

@@ -0,0 +1,58 @@
+const std = @import("std");
+const maxminddb = @import("maxminddb");
+
+const default_db_path: []const u8 = "GeoLite2-City.mmdb";
+
+pub fn main() !void {
+    const allocator = std.heap.smp_allocator;
+
+    const args = try std.process.argsAlloc(allocator);
+    defer std.process.argsFree(allocator, args);
+
+    var db_path: []const u8 = default_db_path;
+    if (args.len > 1) db_path = args[1];
+
+    std.debug.print("Benchmarking with:\n", .{});
+    std.debug.print("  Database: {s}\n", .{db_path});
+    std.debug.print("Opening database...\n", .{});
+
+    var open_timer = try std.time.Timer.start();
+    var db = try maxminddb.Reader.mmap(allocator, db_path, .{});
+    defer db.close();
+    const open_time_ms = @as(f64, @floatFromInt(open_timer.read())) /
+        @as(f64, @floatFromInt(std.time.ns_per_ms));
+    std.debug.print("Database opened successfully in {d} ms. Type: {s}\n", .{
+        open_time_ms,
+        db.metadata.database_type,
+    });
+
+    const network = if (db.metadata.ip_version == 4)
+        maxminddb.Network.all_ipv4
+    else
+        maxminddb.Network.all_ipv6;
+
+    std.debug.print("Starting benchmark...\n", .{});
+    var timer = try std.time.Timer.start();
+
+    var it = try db.scan(maxminddb.any.Value, allocator, network, .{});
+
+    var n: usize = 0;
+    while (try it.next()) |item| {
+        n += 1;
+        item.deinit();
+    }
+
+    const elapsed_ns = timer.read();
+    const elapsed_s = @as(f64, @floatFromInt(elapsed_ns)) /
+        @as(f64, @floatFromInt(std.time.ns_per_s));
+
+    const records_per_second = if (elapsed_s > 0)
+        @as(f64, @floatFromInt(n)) / elapsed_s
+    else
+        0.0;
+
+    std.debug.print("\n--- Benchmark Finished ---\n", .{});
+    std.debug.print("Records:            {d}\n", .{n});
+    std.debug.print("Elapsed Time:       {d} s\n", .{elapsed_s});
+    std.debug.print("Records Per Second: {d}\n", .{records_per_second});
+}