const std = @import("std"); const aws = @import("aws.zig"); const json = @import("json.zig"); var verbose: u8 = 0; pub fn log( comptime level: std.log.Level, comptime scope: @TypeOf(.EnumLiteral), comptime format: []const u8, args: anytype, ) void { // Ignore aws_signing messages if (verbose < 3 and scope == .aws_signing and @intFromEnum(level) >= @intFromEnum(std.log.Level.debug)) return; // Ignore aws_credentials messages if (verbose < 3 and scope == .aws_credentials and @intFromEnum(level) >= @intFromEnum(std.log.Level.debug)) return; // Ignore xml_shaper messages if (verbose < 3 and scope == .xml_shaper and @intFromEnum(level) >= @intFromEnum(std.log.Level.debug)) return; // Ignore date messages if (verbose < 3 and scope == .date and @intFromEnum(level) >= @intFromEnum(std.log.Level.debug)) return; // Ignore awshttp messages if (verbose < 2 and scope == .awshttp and @intFromEnum(level) >= @intFromEnum(std.log.Level.debug)) return; if (verbose < 1 and scope == .aws and @intFromEnum(level) >= @intFromEnum(std.log.Level.debug)) return; const scope_prefix = "(" ++ @tagName(scope) ++ "): "; const prefix = "[" ++ @tagName(level) ++ "] " ++ scope_prefix; // Print the message to stderr, silently ignoring any errors std.debug.getStderrMutex().lock(); defer std.debug.getStderrMutex().unlock(); const stderr = std.io.getStdErr().writer(); nosuspend stderr.print(prefix ++ format ++ "\n", args) catch return; } const Tests = enum { query_no_input, query_with_input, ec2_query_no_input, ec2_query_with_input, json_1_0_query_with_input, json_1_0_query_no_input, json_1_1_query_with_input, json_1_1_query_no_input, rest_json_1_query_no_input, rest_json_1_query_with_input, rest_json_1_work_with_lambda, rest_xml_no_input, rest_xml_anything_but_s3, rest_xml_work_with_s3, }; pub fn main() anyerror!void { var gpa = std.heap.GeneralPurposeAllocator(.{}){}; defer _ = gpa.deinit(); const allocator = gpa.allocator(); var tests = std.ArrayList(Tests).init(allocator); defer tests.deinit(); var args = std.process.args(); while (args.next()) |arg| { if (std.mem.eql(u8, "-v", arg)) { verbose += 1; continue; } inline for (@typeInfo(Tests).Enum.fields) |f| { if (std.mem.eql(u8, f.name, arg)) { try tests.append(@field(Tests, f.name)); break; } } } if (tests.items.len == 0) { inline for (@typeInfo(Tests).Enum.fields) |f| try tests.append(@field(Tests, f.name)); } std.log.info("Start\n", .{}); var client = try aws.Client.init(allocator, .{}); const options = aws.Options{ .region = "us-west-2", .client = client, }; defer client.deinit(); const services = aws.Services(.{ .sts, .ec2, .dynamo_db, .ecs, .lambda, .sqs, .s3, .cloudfront }){}; for (tests.items) |t| { std.log.info("===== Start Test: {s} =====", .{@tagName(t)}); switch (t) { .query_no_input => { const call = try aws.Request(services.sts.get_caller_identity).call(.{}, options); // const call = try client.call(services.sts.get_caller_identity.Request{}, options); defer call.deinit(); std.log.info("arn: {any}", .{call.response.arn}); std.log.info("id: {any}", .{call.response.user_id}); std.log.info("account: {any}", .{call.response.account}); std.log.info("requestId: {any}", .{call.response_metadata.request_id}); }, .query_with_input => { const call = try client.call(services.sqs.list_queues.Request{ .queue_name_prefix = "s", }, options); defer call.deinit(); std.log.info("request id: {any}", .{call.response_metadata.request_id}); std.log.info("account has queues with prefix 's': {}", .{call.response.queue_urls != null}); }, .json_1_0_query_with_input => { const call = try client.call(services.dynamo_db.list_tables.Request{ .limit = 1, }, options); defer call.deinit(); std.log.info("request id: {any}", .{call.response_metadata.request_id}); std.log.info("account has tables: {}", .{call.response.table_names.?.len > 0}); }, .json_1_0_query_no_input => { const call = try client.call(services.dynamo_db.describe_limits.Request{}, options); defer call.deinit(); std.log.info("account read capacity limit: {?d}", .{call.response.account_max_read_capacity_units}); }, .json_1_1_query_with_input => { const call = try client.call(services.ecs.list_clusters.Request{ .max_results = 1, }, options); defer call.deinit(); std.log.info("request id: {any}", .{call.response_metadata.request_id}); std.log.info("account has clusters: {}", .{call.response.cluster_arns.?.len > 0}); }, .json_1_1_query_no_input => { const call = try client.call(services.ecs.list_clusters.Request{}, options); defer call.deinit(); std.log.info("request id: {any}", .{call.response_metadata.request_id}); std.log.info("account has clusters: {}", .{call.response.cluster_arns.?.len > 0}); }, .rest_json_1_query_with_input => { const call = try client.call(services.lambda.list_functions.Request{ .max_items = 1, }, options); defer call.deinit(); std.log.info("request id: {any}", .{call.response_metadata.request_id}); std.log.info("account has functions: {}", .{call.response.functions.?.len > 0}); }, .rest_json_1_query_no_input => { const call = try client.call(services.lambda.list_functions.Request{}, options); defer call.deinit(); std.log.info("request id: {any}", .{call.response_metadata.request_id}); std.log.info("account has functions: {}", .{call.response.functions.?.len > 0}); }, .rest_json_1_work_with_lambda => { const call = try client.call(services.lambda.list_functions.Request{}, options); defer call.deinit(); std.log.info("list request id: {any}", .{call.response_metadata.request_id}); if (call.response.functions) |fns| { if (fns.len > 0) { const func = fns[0]; const arn = func.function_arn.?; // This is a bit ugly. Maybe a helper function in the library would help? var tags = try std.ArrayList(@typeInfo(try typeForField(services.lambda.tag_resource.Request, "tags")).Pointer.child).initCapacity(allocator, 1); defer tags.deinit(); tags.appendAssumeCapacity(.{ .key = "Foo", .value = "Bar" }); const req = services.lambda.tag_resource.Request{ .resource = arn, .tags = tags.items }; const addtag = try aws.Request(services.lambda.tag_resource).call(req, options); defer addtag.deinit(); // const addtag = try client.call(services.lambda.tag_resource.Request{ .resource = arn, .tags = &.{.{ .key = "Foo", .value = "Bar" }} }, options); std.log.info("add tag request id: {any}", .{addtag.response_metadata.request_id}); var keys = [_][]const u8{"Foo"}; // Would love to have a way to express this without burning a var here const deletetag = try aws.Request(services.lambda.untag_resource).call(.{ .tag_keys = keys[0..], .resource = arn }, options); defer deletetag.deinit(); std.log.info("delete tag request id: {any}", .{deletetag.response_metadata.request_id}); } else { std.log.err("no functions to work with", .{}); } } else { std.log.err("no functions to work with", .{}); } }, .ec2_query_no_input => { // Describe regions is a simpler request and easier to debug const result = try client.call(services.ec2.describe_regions.Request{}, options); defer result.deinit(); std.log.info("request id: {any}", .{result.response_metadata.request_id}); std.log.info("region count: {d}", .{result.response.regions.?.len}); }, .ec2_query_with_input => { // Describe instances is more interesting const result = try client.call(services.ec2.describe_instances.Request{ .max_results = 6 }, options); defer result.deinit(); std.log.info("reservation count: {d}", .{result.response.reservations.?.len}); var items: usize = 0; for (result.response.reservations.?) |reservation| { items += reservation.instances.?.len; } std.log.info("items count: {d}", .{items}); var next = result.response.next_token; while (next) |next_token| { std.log.info("more results available: fetching again", .{}); const more = try aws.Request(services.ec2.describe_instances) .call(.{ .next_token = next_token, .max_results = 6 }, options); defer more.deinit(); // we could have exactly 6, which means we have a next token(?!) but not // any actual additional data if (more.response.reservations == null) break; std.log.info("reservation count: {d}", .{more.response.reservations.?.len}); var batch_items: usize = 0; for (more.response.reservations.?) |reservation| { batch_items += reservation.instances.?.len; } std.log.info("items count: {d}", .{batch_items}); items += batch_items; std.log.info("total items count: {d}", .{items}); next = more.response.next_token; } }, .rest_xml_no_input => { const result = try client.call(services.s3.list_buckets.Request{}, options); defer result.deinit(); std.log.info("request id: {any}", .{result.response_metadata.request_id}); std.log.info("bucket count: {d}", .{result.response.buckets.?.len}); }, .rest_xml_anything_but_s3 => { const result = try client.call(services.cloudfront.list_key_groups.Request{}, options); defer result.deinit(); std.log.info("request id: {any}", .{result.response_metadata.request_id}); const list = result.response.key_group_list.?; std.log.info("key group list max: {?d}", .{list.max_items}); std.log.info("key group quantity: {d}", .{list.quantity}); }, .rest_xml_work_with_s3 => { const key = "i/am/a/teapot/foo"; // const key = "foo"; const bucket = blk: { const result = try client.call(services.s3.list_buckets.Request{}, options); defer result.deinit(); const bucket = result.response.buckets.?[result.response.buckets.?.len - 1]; std.log.info("ListBuckets request id: {any}", .{result.response_metadata.request_id}); std.log.info("bucket name: {any}", .{bucket.name.?}); break :blk try allocator.dupe(u8, bucket.name.?); }; defer allocator.free(bucket); const location = blk: { const result = try aws.Request(services.s3.get_bucket_location).call(.{ .bucket = bucket, }, options); defer result.deinit(); const location = result.response.location_constraint.?; std.log.info("GetBucketLocation request id: {any}", .{result.response_metadata.request_id}); std.log.info("location: {any}", .{location}); break :blk try allocator.dupe(u8, location); }; defer allocator.free(location); const s3opts = aws.Options{ .region = location, .client = client, }; { const result = try aws.Request(services.s3.put_object).call(.{ .bucket = bucket, .key = key, .content_type = "text/plain", .body = "bar", .storage_class = "STANDARD", }, s3opts); std.log.info("PutObject Request id: {any}", .{result.response_metadata.request_id}); std.log.info("PutObject etag: {any}", .{result.response.e_tag.?}); defer result.deinit(); } { // Note that boto appears to redirect by default, but java // does not. We will not const result = try aws.Request(services.s3.get_object).call(.{ .bucket = bucket, .key = key, }, s3opts); std.log.info("GetObject Request id: {any}", .{result.response_metadata.request_id}); std.log.info("GetObject Body: {any}", .{result.response.body}); std.log.info("GetObject etag: {any}", .{result.response.e_tag.?}); std.log.info("GetObject last modified (seconds since epoch): {d}", .{result.response.last_modified.?}); defer result.deinit(); } { const result = try aws.Request(services.s3.delete_object).call(.{ .bucket = bucket, .key = key, }, s3opts); std.log.info("DeleteObject Request id: {any}", .{result.response_metadata.request_id}); defer result.deinit(); } { const result = try aws.Request(services.s3.list_objects).call(.{ .bucket = bucket, }, s3opts); std.log.info("ListObject Request id: {any}", .{result.response_metadata.request_id}); std.log.info("Object count: {d}", .{result.response.contents.?.len}); defer result.deinit(); } }, } std.log.info("===== End Test: {s} =====\n", .{@tagName(t)}); } // if (test_twice) { // std.time.sleep(1000 * std.time.ns_per_ms); // std.log.info("second request", .{}); // // var client2 = aws.Aws.init(allocator); // defer client2.deinit(); // const resp2 = try client2.call(services.sts.get_caller_identity.Request{}, options); // catch here and try alloc? // defer resp2.deinit(); // } std.log.info("===== Tests complete =====", .{}); } fn typeForField(comptime T: type, comptime field_name: []const u8) !type { const ti = @typeInfo(T); switch (ti) { .Struct => { inline for (ti.Struct.fields) |field| { if (std.mem.eql(u8, field.name, field_name)) return field.type; } }, else => return error.TypeIsNotAStruct, // should not hit this } return error.FieldNotFound; } // TODO: Move into json.zig pub fn jsonFun() !void { // Standard behavior const payload = \\{"GetCallerIdentityResponse":{"GetCallerIdentityResult":{"Account":"0123456789","Arn":"arn:aws:iam::0123456789:user/test","UserId":"MYUSERID"},"ResponseMetadata":{"RequestId":"3b80a99b-7df8-4bcb-96ee-b2759878a5f2"}}} ; const Ret3 = struct { getCallerIdentityResponse: struct { getCallerIdentityResult: struct { account: []u8, arn: []u8, user_id: []u8 }, responseMetadata: struct { requestId: []u8 } }, }; var stream3 = json.TokenStream.init(payload); const res3 = json.parse(Ret3, &stream3, .{ .allocator = std.heap.c_allocator, .allow_camel_case_conversion = true, // new option .allow_snake_case_conversion = true, // new option .allow_unknown_fields = true, // new option }) catch unreachable; std.log.info("{}", .{res3}); std.log.info("{any}", .{res3.getCallerIdentityResponse.getCallerIdentityResult.user_id}); }