Line data Source code
1 : /* The backtest topology exercises offline replay with mocked consensus.
2 :
3 : The smaller topology is:
4 : shred_out replay_execr
5 : backtest-------------->replay------------->execrp
6 : ^ |^ | ^ |
7 : |____________________|| | |________________|
8 : replay_out | | execrp_replay
9 : | |------------------------------>no consumer
10 : no producer------------- stake_out, txsend_out, poh_out
11 : store_replay
12 :
13 : */
14 : #define _GNU_SOURCE
15 : #include "../../firedancer/topology.h"
16 : #include "../../shared/commands/configure/configure.h"
17 : #include "../../shared/commands/run/run.h" /* initialize_workspaces */
18 : #include "../../shared/commands/watch/watch.h"
19 : #include "../../shared/fd_config.h" /* config_t */
20 : #include "../../shared/fd_config_json.h"
21 : #include "../../../disco/topo/fd_topob.h"
22 : #include "../../../disco/topo/fd_cpu_topo.h"
23 : #include "../../../util/pod/fd_pod_format.h"
24 : #include "../../../disco/gui/fd_gui_config_parse.h"
25 : #include "../../../disco/diag/fd_diag_tile.h"
26 : #include "../../../ballet/base58/fd_base58.h"
27 : #include "../../../disco/genesis/fd_genesis_cluster.h"
28 : #include "../../../discof/genesis/fd_genesi_tile.h"
29 : #include "../../../discof/genesis/genesis_hash.h"
30 : #include "../../../discof/replay/fd_replay_tile.h"
31 : #include "../../../discof/restore/fd_snapct_tile.h"
32 : #include "../../../discof/restore/utils/fd_ssctrl.h"
33 : #include "../../../discof/restore/utils/fd_ssmsg.h"
34 : #include "../../../discof/tower/fd_tower_tile.h"
35 : #include "../../../discof/votor/fd_votor_tile.h"
36 : #include "../../../discof/replay/fd_execrp.h"
37 : #include "../../../disco/shred/fd_shred_tile.h"
38 : #include "../../../discof/repair/fd_repair_tile.h"
39 : #include "../../../flamenco/capture/fd_capture_ctx.h"
40 : #include "../../../disco/pack/fd_pack_cost.h"
41 : #include "../../../flamenco/runtime/fd_cost_tracker.h"
42 :
43 : #include <errno.h>
44 : #include <stdlib.h> /* getenv */
45 : #include <unistd.h>
46 : #include <fcntl.h>
47 :
48 : extern fd_topo_obj_callbacks_t * CALLBACKS[];
49 : fd_topo_run_tile_t fdctl_tile_run( fd_topo_tile_t const * tile );
50 :
51 : void
52 : backtest_cmd_args( int * pargc,
53 : char *** pargv,
54 0 : args_t * args ) {
55 0 : args->backtest.no_watch = fd_env_strip_cmdline_contains( pargc, pargv, "--no-watch" );
56 0 : }
57 :
58 : static void
59 0 : backtest_topo( config_t * config ) {
60 :
61 0 : config->development.sandbox = 0;
62 0 : config->development.no_clone = 1;
63 :
64 0 : config->firedancer.layout.resolv_tile_count = 0;
65 :
66 0 : ulong execrp_tile_cnt = config->firedancer.layout.execrp_tile_count;
67 0 : ulong snapdc_tile_cnt = config->firedancer.layout.snapdc_tile_count;
68 :
69 0 : int disable_snap_loader = !config->gossip.entrypoints_cnt;
70 0 : int solcap_enabled = strlen( config->capture.solcap_capture )>0;
71 0 : int telemetry_enabled = config->telemetry && strcmp( config->tiles.event.url, "" );
72 :
73 0 : fd_topo_t * topo = { fd_topob_new( &config->topo, config->name ) };
74 0 : topo->max_page_size = fd_cstr_to_shmem_page_sz( config->hugetlbfs.max_page_size );
75 0 : topo->gigantic_page_threshold = config->hugetlbfs.gigantic_page_threshold_mib << 20;
76 :
77 0 : ushort parsed_tile_to_cpu[ FD_TILE_MAX ];
78 : /* Unassigned tiles will be floating, unless auto topology is enabled. */
79 0 : for( ulong i=0UL; i<FD_TILE_MAX; i++ ) parsed_tile_to_cpu[ i ] = USHORT_MAX;
80 :
81 0 : int is_auto_affinity = !strcmp( config->layout.affinity, "auto" );
82 :
83 0 : fd_topo_cpus_t cpus[1];
84 0 : fd_topo_cpus_init( cpus );
85 :
86 0 : ulong affinity_tile_cnt = 0UL;
87 0 : if( FD_LIKELY( !is_auto_affinity ) ) affinity_tile_cnt = fd_topob_parse_affinity_cstr( config->layout.affinity, parsed_tile_to_cpu, 1, 1 );
88 :
89 0 : ulong tile_to_cpu[ FD_TILE_MAX ] = {0};
90 0 : for( ulong i=0UL; i<affinity_tile_cnt; i++ ) {
91 0 : ushort cpu = (ushort)( parsed_tile_to_cpu[ i ] & ~FD_TOPOB_CPU_SHARED );
92 0 : if( FD_UNLIKELY( parsed_tile_to_cpu[ i ]!=USHORT_MAX && cpu>=cpus->cpu_cnt ) )
93 0 : FD_LOG_ERR(( "[layout.affinity] specifies CPU %hu but the system only has %lu CPUs", cpu, cpus->cpu_cnt ));
94 0 : tile_to_cpu[ i ] = fd_ulong_if( parsed_tile_to_cpu[ i ]==USHORT_MAX, ULONG_MAX, (ulong)parsed_tile_to_cpu[ i ] );
95 0 : }
96 :
97 0 : ulong cpu_idx = 1;
98 0 : # define NEXT_CPU ( is_auto_affinity ? cpu_idx++ : tile_to_cpu[ topo->tile_cnt ] )
99 0 : # define FLOAT_CPU ( is_auto_affinity ? ULONG_MAX : tile_to_cpu[ topo->tile_cnt ] )
100 :
101 0 : fd_topob_wksp( topo, "metric" );
102 0 : fd_topob_wksp( topo, "metric_in" );
103 0 : fd_topob_tile( topo, "metric", "metric", "metric_in", FLOAT_CPU, 0, 0, 0, 1 );
104 :
105 0 : fd_topob_wksp( topo, "backt" );
106 0 : fd_topo_tile_t * backt_tile = fd_topob_tile( topo, "backt", "backt", "metric_in", NEXT_CPU, 0, 0, 0, 0 );
107 :
108 0 : fd_topob_wksp( topo, "replay" );
109 0 : fd_topo_tile_t * replay_tile = fd_topob_tile( topo, "replay", "replay", "metric_in", NEXT_CPU, 0, 1, 0, 0 );
110 :
111 0 : fd_topo_obj_t * node_info_obj = fd_topob_obj( topo, "node_info", "replay" );
112 0 : fd_topob_tile_uses( topo, replay_tile, node_info_obj, FD_SHMEM_JOIN_MODE_READ_WRITE );
113 0 : FD_TEST( fd_pod_insertf_ulong( topo->props, node_info_obj->id, "node_info" ) );
114 :
115 0 : fd_topob_wksp( topo, "accdb" )->core_dump_level = FD_TOPO_CORE_DUMP_LEVEL_FULL;
116 0 : fd_topo_tile_t * accdb_tile = fd_topob_tile( topo, "accdb", "accdb", "metric_in", NEXT_CPU, 0, 0, 0, 0 );
117 :
118 0 : fd_topob_wksp( topo, "progcache" );
119 0 : setup_topo_progcache( topo, "progcache",
120 0 : config->firedancer.runtime.max_live_slots,
121 0 : config->firedancer.runtime.program_cache_size_mib<<20 );
122 0 : ulong progcache_obj_id; FD_TEST( (progcache_obj_id = fd_pod_query_ulong( topo->props, "progcache", ULONG_MAX ) )!=ULONG_MAX );
123 0 : fd_topob_tile_uses( topo, replay_tile, &topo->objs[ progcache_obj_id ], FD_SHMEM_JOIN_MODE_READ_WRITE );
124 0 : fd_topob_tile_uses( topo, accdb_tile, &topo->objs[ progcache_obj_id ], FD_SHMEM_JOIN_MODE_READ_WRITE );
125 :
126 0 : fd_topob_wksp( topo, "accdb_data" );
127 0 : fd_topo_obj_t * accdb_obj = setup_topo_accdb( topo, "accdb_data",
128 0 : config->firedancer.accounts.max_accounts,
129 0 : config->firedancer.runtime.max_live_slots,
130 0 : FD_RUNTIME_MAX_ACC_WRITES_PER_SLOT,
131 0 : 8192UL,
132 0 : 1UL<<35UL,
133 0 : config->firedancer.accounts.cache_size_gib*(1UL<<30UL),
134 0 : config->tiles.bundle.enabled,
135 0 : execrp_tile_cnt+3UL,
136 0 : 0UL );
137 0 : FD_TEST( fd_pod_insertf_ulong( topo->props, accdb_obj->id, "accdb" ) );
138 :
139 : /**********************************************************************/
140 : /* Add the executor tiles to topo */
141 : /**********************************************************************/
142 0 : fd_topob_wksp( topo, "execrp" );
143 0 : #define FOR(cnt) for( ulong i=0UL; i<cnt; i++ )
144 0 : FOR(execrp_tile_cnt) fd_topob_tile( topo, "execrp", "execrp", "metric_in", NEXT_CPU, 0, 0, 0, 0 );
145 :
146 : /**********************************************************************/
147 : /* Add the capture tile to topo */
148 : /**********************************************************************/
149 0 : if( solcap_enabled ) {
150 0 : fd_topob_wksp( topo, "solcap" );
151 0 : fd_topob_tile( topo, "solcap", "solcap", "metric_in", NEXT_CPU, 0, 0, 0, 0 );
152 0 : }
153 :
154 0 : fd_topob_wksp( topo, "diag" );
155 0 : fd_topob_tile( topo, "diag", "diag", "metric_in", FLOAT_CPU, 0, 0, 0, 0 );
156 :
157 : /**********************************************************************/
158 : /* Add the snapshot tiles to topo */
159 : /**********************************************************************/
160 0 : fd_topo_tile_t * snapin_tile = NULL;
161 0 : if( FD_UNLIKELY( !disable_snap_loader ) ) {
162 0 : fd_topob_wksp( topo, "snapct" );
163 0 : fd_topob_wksp( topo, "snapld" );
164 0 : fd_topob_wksp( topo, "snapdc" );
165 0 : fd_topob_wksp( topo, "snapin" );
166 0 : fd_topob_wksp( topo, "snapwr" );
167 :
168 0 : fd_topo_tile_t * snapct_tile = fd_topob_tile( topo, "snapct", "snapct", "metric_in", NEXT_CPU, 0, 0, 0, 0 );
169 0 : fd_topo_tile_t * snapld_tile = fd_topob_tile( topo, "snapld", "snapld", "metric_in", NEXT_CPU, 0, 0, 0, 0 );
170 0 : FOR(snapdc_tile_cnt) fd_topob_tile( topo, "snapdc", "snapdc", "metric_in", NEXT_CPU, 0, 0, 0, 0 )->allow_shutdown = 1;
171 0 : snapin_tile = fd_topob_tile( topo, "snapin", "snapin", "metric_in", NEXT_CPU, 0, 0, 0, 0 );
172 0 : fd_topo_tile_t * snapwr_tile = fd_topob_tile( topo, "snapwr", "snapwr", "metric_in", NEXT_CPU, 0, 0, 0, 0 );
173 :
174 0 : snapct_tile->allow_shutdown = 1;
175 0 : snapld_tile->allow_shutdown = 1;
176 0 : snapin_tile->allow_shutdown = 1;
177 0 : snapwr_tile->allow_shutdown = 1;
178 0 : } else {
179 0 : fd_topob_wksp( topo, "genesi" );
180 0 : fd_topob_tile( topo, "genesi", "genesi", "metric_in", NEXT_CPU, 0, 0, 0, 0 )->allow_shutdown = 1;
181 0 : }
182 :
183 : /**********************************************************************/
184 : /* Add the event + sign tiles to topo */
185 : /**********************************************************************/
186 0 : if( FD_LIKELY( telemetry_enabled ) ) {
187 0 : fd_topob_wksp( topo, "sign" );
188 0 : fd_topob_wksp( topo, "event" );
189 0 : fd_topob_wksp( topo, "event_sign" );
190 0 : fd_topob_wksp( topo, "sign_event" );
191 :
192 0 : fd_topob_tile( topo, "sign", "sign", "metric_in", NEXT_CPU, 0, 1, 1, 0 );
193 0 : fd_topo_tile_t * event_tile = fd_topob_tile( topo, "event", "event", "metric_in", NEXT_CPU, 0, 1, 0, 1 );
194 :
195 0 : ushort shred_version = 0;
196 0 : uchar genesis_hash[ 32 ] = {0};
197 0 : if( FD_UNLIKELY( -1==read_genesis_bin( config->paths.genesis, &shred_version, genesis_hash ) ) ) {
198 0 : FD_LOG_ERR(( "could not read genesis `%s` for the event tile (%i-%s)",
199 0 : config->paths.genesis, errno, fd_io_strerror( errno ) ));
200 0 : }
201 0 : fd_memcpy( event_tile->event.genesis_hash, genesis_hash, 32UL );
202 0 : event_tile->event.shred_version = shred_version;
203 :
204 0 : topo->resolved_config_json_len = fd_config_to_json( config, topo->resolved_config_json, sizeof(topo->resolved_config_json) );
205 0 : topo->user_config_json_len = fd_config_user_toml_to_json( config, topo->user_config_json, sizeof(topo->user_config_json) );
206 :
207 0 : fd_topob_link( topo, "event_sign", "event_sign", 128UL, 317UL, 1UL );
208 0 : fd_topob_link( topo, "sign_event", "sign_event", 128UL, 64UL, 1UL );
209 :
210 0 : fd_topob_tile_in ( topo, "sign", 0UL, "metric_in", "event_sign", 0UL, FD_TOPOB_UNRELIABLE, FD_TOPOB_POLLED );
211 0 : fd_topob_tile_out( topo, "event", 0UL, "event_sign", 0UL );
212 0 : fd_topob_tile_in ( topo, "event", 0UL, "metric_in", "sign_event", 0UL, FD_TOPOB_UNRELIABLE, FD_TOPOB_UNPOLLED );
213 0 : fd_topob_tile_out( topo, "sign", 0UL, "sign_event", 0UL );
214 0 : }
215 :
216 : /**********************************************************************/
217 : /* Setup backtest->replay link (repair_out) in topo */
218 : /**********************************************************************/
219 :
220 : /* The repair tile is replaced by the backtest tile for the repair to
221 : replay link. The frag interface is a "slice", ie. entry batch,
222 : which is provided by the backtest tile, which reads in the entry
223 : batches from the CLI-specified source (eg. pcap/pcapng file). */
224 :
225 0 : fd_topob_wksp( topo, "repair_out" );
226 0 : fd_topob_link( topo, "repair_out", "repair_out", 65536UL, sizeof(fd_repair_fec_complete_t), 1UL );
227 0 : fd_topob_tile_in( topo, "replay", 0UL, "metric_in", "repair_out", 0UL, FD_TOPOB_RELIABLE, FD_TOPOB_POLLED );
228 0 : fd_topob_tile_out( topo, "backt", 0UL, "repair_out", 0UL );
229 :
230 : /**********************************************************************/
231 : /* Setup snapshot links in topo */
232 : /**********************************************************************/
233 0 : if( FD_LIKELY( !disable_snap_loader ) ) {
234 0 : fd_topob_wksp( topo, "snapct_ld" );
235 0 : fd_topob_wksp( topo, "snapld_dc" );
236 0 : fd_topob_wksp( topo, "snapdc_in" );
237 :
238 0 : fd_topob_wksp( topo, "snapin_manif" );
239 0 : fd_topob_wksp( topo, "snapct_repr" );
240 :
241 0 : fd_topob_wksp( topo, "snapin_ct" );
242 0 : fd_topob_wksp( topo, "snapwr_ct" );
243 :
244 0 : fd_topob_link( topo, "snapct_ld", "snapct_ld", 128UL, sizeof(fd_ssctrl_init_t), 1UL );
245 0 : fd_topob_link( topo, "snapld_dc", "snapld_dc", FD_SNAPSHOT_DATA_DEPTH, FD_SNAPSHOT_DATA_MTU, 1UL );
246 0 : FOR(snapdc_tile_cnt) fd_topob_link( topo, "snapdc_in", "snapdc_in", FD_SNAPSHOT_DATA_DEPTH, FD_SNAPSHOT_DATA_MTU, 1UL );
247 :
248 0 : fd_topob_link( topo, "snapin_manif", "snapin_manif", 4UL, sizeof(fd_snapshot_manifest_t), 1UL ); /* TODO: Should be depth 1 or 2 but replay backpressures */
249 0 : fd_topob_link( topo, "snapct_repr", "snapct_repr", 128UL, 0UL, 1UL )->permit_no_consumers = 1;
250 :
251 0 : fd_topob_link( topo, "snapin_ct", "snapin_ct", 128UL, 0UL, 1UL );
252 0 : fd_topob_link( topo, "snapwr_ct", "snapwr_ct", 128UL, 0UL, 1UL );
253 :
254 0 : fd_topob_tile_in ( topo, "snapct", 0UL, "metric_in", "snapin_ct", 0UL, FD_TOPOB_RELIABLE, FD_TOPOB_POLLED );
255 0 : fd_topob_tile_in ( topo, "snapct", 0UL, "metric_in", "snapwr_ct", 0UL, FD_TOPOB_RELIABLE, FD_TOPOB_POLLED );
256 0 : fd_topob_tile_in ( topo, "snapct", 0UL, "metric_in", "snapld_dc", 0UL, FD_TOPOB_RELIABLE, FD_TOPOB_POLLED );
257 0 : fd_topob_tile_out( topo, "snapct", 0UL, "snapct_ld", 0UL );
258 0 : fd_topob_tile_out( topo, "snapct", 0UL, "snapct_repr", 0UL );
259 0 : fd_topob_tile_in ( topo, "snapld", 0UL, "metric_in", "snapct_ld", 0UL, FD_TOPOB_RELIABLE, FD_TOPOB_POLLED );
260 0 : fd_topob_tile_out( topo, "snapld", 0UL, "snapld_dc", 0UL );
261 0 : FOR(snapdc_tile_cnt) fd_topob_tile_in ( topo, "snapdc", i, "metric_in", "snapld_dc", 0UL, FD_TOPOB_RELIABLE, FD_TOPOB_POLLED );
262 0 : FOR(snapdc_tile_cnt) fd_topob_tile_out( topo, "snapdc", i, "snapdc_in", i );
263 0 : FOR(snapdc_tile_cnt) fd_topob_tile_in ( topo, "snapin", 0UL, "metric_in", "snapdc_in", i, FD_TOPOB_RELIABLE, FD_TOPOB_POLLED );
264 0 : FOR(snapdc_tile_cnt) fd_topob_tile_in ( topo, "snapwr", 0UL, "metric_in", "snapdc_in", i, FD_TOPOB_RELIABLE, FD_TOPOB_POLLED );
265 :
266 0 : fd_topob_tile_out( topo, "snapin", 0UL, "snapin_manif", 0UL );
267 0 : fd_topob_tile_in ( topo, "replay", 0UL, "metric_in", "snapin_manif", 0UL, FD_TOPOB_RELIABLE, FD_TOPOB_POLLED );
268 :
269 0 : fd_topob_tile_out( topo, "snapin", 0UL, "snapin_ct", 0UL );
270 0 : fd_topob_tile_out( topo, "snapwr", 0UL, "snapwr_ct", 0UL );
271 0 : } else {
272 0 : fd_topob_wksp( topo, "genesi_out" );
273 0 : fd_topob_link( topo, "genesi_out", "genesi_out", 1UL,
274 0 : fd_genesi_tile_mtu( config->firedancer.development.genesis.max_file_size_mib<<20 ), 1UL );
275 0 : fd_topob_tile_out( topo, "genesi", 0UL, "genesi_out", 0UL );
276 0 : fd_topob_tile_in ( topo, "replay", 0UL, "metric_in", "genesi_out", 0UL, FD_TOPOB_RELIABLE, FD_TOPOB_POLLED );
277 0 : }
278 :
279 : /**********************************************************************/
280 : /* More backtest->replay links in topo */
281 : /**********************************************************************/
282 :
283 : /* The tower tile is replaced by the backtest tile for the tower to
284 : replay link. The backtest tile simply sends monotonically
285 : increasing rooted slot numbers to the replay tile, once after each
286 : "replayed a full slot" notification received from the replay tile.
287 : This allows the replay tile to advance its watermark, and publish
288 : various data structures. This is an oversimplified barebones mock
289 : of the tower tile. */
290 0 : fd_topob_wksp( topo, "tower_out" );
291 0 : fd_topob_link( topo, "tower_out", "tower_out", 1024UL, sizeof(fd_tower_slot_done_t), 1UL );
292 0 : fd_topob_tile_in( topo, "replay", 0UL, "metric_in", "tower_out", 0UL, FD_TOPOB_RELIABLE, FD_TOPOB_POLLED );
293 0 : fd_topob_tile_out( topo, "backt", 0UL, "tower_out", 0UL );
294 :
295 0 : if( FD_UNLIKELY( config->firedancer.development.alpenglow ) ) {
296 0 : fd_topob_wksp( topo, "votor_out" );
297 0 : fd_topob_link( topo, "votor_out", "votor_out", 1024UL, sizeof(fd_votor_msg_t), 2UL );
298 0 : fd_topob_tile_in( topo, "replay", 0UL, "metric_in", "votor_out", 0UL, FD_TOPOB_RELIABLE, FD_TOPOB_POLLED );
299 0 : fd_topob_tile_out( topo, "backt", 0UL, "votor_out", 0UL );
300 0 : }
301 :
302 : /**********************************************************************/
303 : /* Setup replay->stake/send/poh links in topo w/o consumers */
304 : /**********************************************************************/
305 0 : fd_topob_wksp( topo, "replay_epoch" );
306 0 : fd_topob_link( topo, "replay_epoch", "replay_epoch", 16UL, FD_EPOCH_OUT_MTU, 1UL );
307 0 : fd_topob_tile_out( topo, "replay", 0UL, "replay_epoch", 0UL );
308 0 : topo->links[ replay_tile->out_link_id[ fd_topo_find_tile_out_link( topo, replay_tile, "replay_epoch", 0 ) ] ].permit_no_consumers = 1;
309 :
310 : /**********************************************************************/
311 : /* Setup replay->backtest link (replay_notif) in topo */
312 : /**********************************************************************/
313 :
314 0 : fd_topob_wksp( topo, "replay_out" );
315 0 : fd_topob_link( topo, "replay_out", "replay_out", 8192UL, sizeof( fd_replay_message_t ), 1UL );
316 0 : fd_topob_tile_out( topo, "replay", 0UL, "replay_out", 0UL );
317 0 : fd_topob_tile_in ( topo, "backt", 0UL, "metric_in", "replay_out", 0UL, FD_TOPOB_RELIABLE, FD_TOPOB_POLLED );
318 0 : if( FD_LIKELY( !disable_snap_loader ) ) {
319 0 : fd_topob_tile_in ( topo, "backt", 0UL, "metric_in", "snapin_manif", 0UL, FD_TOPOB_RELIABLE, FD_TOPOB_POLLED );
320 0 : } else {
321 0 : fd_topob_tile_in ( topo, "backt", 0UL, "metric_in", "genesi_out", 0UL, FD_TOPOB_RELIABLE, FD_TOPOB_POLLED );
322 0 : }
323 :
324 : /**********************************************************************/
325 : /* Setup replay->exec link in topo */
326 : /**********************************************************************/
327 0 : fd_topob_wksp( topo, "replay_execrp" );
328 0 : fd_topob_link( topo, "replay_execrp", "replay_execrp", 16384UL, sizeof(fd_execrp_task_msg_t), 1UL );
329 0 : fd_topob_tile_out( topo, "replay", 0UL, "replay_execrp", 0UL );
330 0 : for( ulong i=0UL; i<execrp_tile_cnt; i++ ) {
331 0 : fd_topob_tile_in( topo, "execrp", i, "metric_in", "replay_execrp", 0UL, FD_TOPOB_RELIABLE, FD_TOPOB_POLLED );
332 0 : }
333 :
334 : /**********************************************************************/
335 : /* Setup exec->replay links in topo, to send solcap account updates
336 : so that they are serialized, and to notify replay tile that a txn
337 : has been finalized by the exec tile. */
338 : /**********************************************************************/
339 0 : fd_topob_wksp( topo, "execrp_replay" );
340 :
341 0 : FOR(execrp_tile_cnt) fd_topob_link( topo, "execrp_replay", "execrp_replay", 16384UL, sizeof(fd_execrp_task_done_msg_t), 1UL );
342 :
343 0 : FOR(execrp_tile_cnt) fd_topob_tile_out( topo, "execrp", i, "execrp_replay", i );
344 0 : FOR(execrp_tile_cnt) fd_topob_tile_in( topo, "replay", 0UL, "metric_in", "execrp_replay", i, FD_TOPOB_RELIABLE, FD_TOPOB_POLLED );
345 :
346 : /**********************************************************************/
347 : /* Add the gui tile to topo */
348 : /**********************************************************************/
349 :
350 : /* The gui tile consumes the subset of its usual inputs that exist in
351 : the backtest topology (there is no networking, gossip, shred or
352 : leader pipeline here). This is enough for the frontend to show
353 : snapshot load and replay progress. Genesis-mode backtest (no
354 : snapshot loader) is not supported: the gui boot-progress sampler
355 : requires the snapshot tiles. */
356 0 : if( FD_UNLIKELY( config->tiles.gui.enabled && !disable_snap_loader ) ) {
357 0 : fd_topob_wksp( topo, "gui" );
358 0 : fd_topo_tile_t * gui_tile = fd_topob_tile( topo, "gui", "gui", "metric_in", NEXT_CPU, 0, 1, 0, 1 );
359 :
360 0 : fd_topob_wksp( topo, "diag_gui" );
361 0 : fd_topob_link( topo, "diag_gui", "diag_gui", 4UL, sizeof(fd_diag_system_resources_t), 1UL );
362 0 : fd_topob_tile_out( topo, "diag", 0UL, "diag_gui", 0UL );
363 0 : fd_topob_tile_in( topo, "gui", 0UL, "metric_in", "diag_gui", 0UL, FD_TOPOB_UNRELIABLE, FD_TOPOB_POLLED );
364 :
365 0 : fd_topob_wksp( topo, "snapct_gui" );
366 0 : fd_topob_wksp( topo, "snapin_gui" );
367 0 : fd_topob_link( topo, "snapct_gui", "snapct_gui", 128UL, sizeof(fd_snapct_update_t), 1UL );
368 0 : fd_topob_link( topo, "snapin_gui", "snapin_gui", 128UL, FD_GUI_CONFIG_PARSE_MAX_VALID_ACCT_SZ, 1UL );
369 0 : fd_topob_tile_out( topo, "snapct", 0UL, "snapct_gui", 0UL );
370 0 : fd_topob_tile_out( topo, "snapin", 0UL, "snapin_gui", 0UL );
371 :
372 0 : /**/ fd_topob_tile_in( topo, "gui", 0UL, "metric_in", "tower_out", 0UL, FD_TOPOB_RELIABLE, FD_TOPOB_POLLED );
373 0 : /**/ fd_topob_tile_in( topo, "gui", 0UL, "metric_in", "replay_out", 0UL, FD_TOPOB_RELIABLE, FD_TOPOB_POLLED );
374 0 : /**/ fd_topob_tile_in( topo, "gui", 0UL, "metric_in", "replay_epoch", 0UL, FD_TOPOB_RELIABLE, FD_TOPOB_POLLED );
375 0 : FOR(execrp_tile_cnt) fd_topob_tile_in( topo, "gui", 0UL, "metric_in", "execrp_replay", i, FD_TOPOB_RELIABLE, FD_TOPOB_POLLED );
376 0 : /**/ fd_topob_tile_in( topo, "gui", 0UL, "metric_in", "snapct_gui", 0UL, FD_TOPOB_RELIABLE, FD_TOPOB_POLLED );
377 0 : /**/ fd_topob_tile_in( topo, "gui", 0UL, "metric_in", "snapin_gui", 0UL, FD_TOPOB_RELIABLE, FD_TOPOB_POLLED );
378 0 : /**/ fd_topob_tile_in( topo, "gui", 0UL, "metric_in", "snapin_manif", 0UL, FD_TOPOB_RELIABLE, FD_TOPOB_POLLED );
379 :
380 0 : fd_topob_tile_uses( topo, gui_tile, accdb_obj, FD_SHMEM_JOIN_MODE_READ_ONLY );
381 0 : }
382 :
383 : /**********************************************************************/
384 : /* Setup the shared objs used by replay and exec tiles */
385 : /**********************************************************************/
386 :
387 0 : if( FD_UNLIKELY( solcap_enabled ) ) {
388 : /* 32 sections of SOLCAP_WRITE_ACCOUNT_DATA_MTU bytes each ≈ 4MB.
389 : This is done to ideally avoid cache thrashing and allow for all
390 : the links to sit on L3 cache. */
391 0 : fd_topob_link( topo, "cap_repl", "solcap", 32UL, SOLCAP_WRITE_ACCOUNT_DATA_MTU, 1UL );
392 0 : fd_topob_tile_out( topo, "replay", 0UL, "cap_repl", 0UL );
393 0 : fd_topob_tile_in( topo, "solcap", 0UL, "metric_in", "cap_repl", 0UL, FD_TOPOB_RELIABLE, FD_TOPOB_POLLED );
394 0 : FOR(execrp_tile_cnt) fd_topob_link( topo, "cap_execrp", "solcap", 32UL, SOLCAP_WRITE_ACCOUNT_DATA_MTU, 1UL );
395 0 : FOR(execrp_tile_cnt) fd_topob_tile_out( topo, "execrp", i, "cap_execrp", i );
396 0 : FOR(execrp_tile_cnt) fd_topob_tile_in( topo, "solcap", 0UL, "metric_in", "cap_execrp", i, FD_TOPOB_RELIABLE, FD_TOPOB_POLLED );
397 0 : }
398 :
399 0 : fd_topob_wksp( topo, "store" );
400 0 : fd_topo_link_t const * repair_out_link = &topo->links[ fd_topo_find_link( topo, "repair_out", 0UL ) ];
401 0 : ulong fec_sets_per_slot = fd_ulong_if( config->firedancer.development.fixed_fec_sets,
402 0 : config->limits.max_shreds_per_block/FD_FEC_SHRED_CNT, config->limits.max_shreds_per_block );
403 0 : ulong store_fec_max = config->firedancer.runtime.max_live_slots * fec_sets_per_slot + repair_out_link->depth + 1UL;
404 0 : ulong store_fec_data_max = fd_ulong_if( config->firedancer.development.fixed_fec_sets, 31840UL, 63985UL );
405 0 : fd_topo_obj_t * store_obj = setup_topo_store( topo, "store", store_fec_max, store_fec_data_max, 0UL, config->tiles.shred.shred_cache_size_mib, 0UL, config->limits.max_shreds_per_block, config->paths.shredb );
406 0 : fd_topob_tile_uses( topo, backt_tile, store_obj, FD_SHMEM_JOIN_MODE_READ_WRITE );
407 0 : fd_topob_tile_uses( topo, replay_tile, store_obj, FD_SHMEM_JOIN_MODE_READ_WRITE );
408 0 : FD_TEST( fd_pod_insertf_ulong( topo->props, store_obj->id, "store" ) );
409 :
410 0 : fd_topob_wksp( topo, "banks" );
411 0 : fd_topo_obj_t * banks_obj = setup_topo_banks( topo, "banks", config->firedancer.runtime.max_live_slots, config->firedancer.runtime.max_fork_width, config->development.bench.max_cost_per_block );
412 0 : fd_topob_tile_uses( topo, replay_tile, banks_obj, FD_SHMEM_JOIN_MODE_READ_WRITE );
413 0 : FOR(execrp_tile_cnt) fd_topob_tile_uses( topo, &topo->tiles[ fd_topo_find_tile( topo, "execrp", i ) ], banks_obj, FD_SHMEM_JOIN_MODE_READ_WRITE );
414 0 : FD_TEST( fd_pod_insertf_ulong( topo->props, banks_obj->id, "banks" ) );
415 :
416 0 : fd_topob_wksp( topo, "txncache" );
417 0 : fd_topo_obj_t * txncache_obj = setup_topo_txncache( topo, "txncache", config->firedancer.runtime.max_live_slots, 2UL*config->limits.max_txn_per_slot );
418 0 : fd_topob_tile_uses( topo, replay_tile, txncache_obj, FD_SHMEM_JOIN_MODE_READ_WRITE );
419 0 : if( FD_LIKELY( !disable_snap_loader ) ) fd_topob_tile_uses( topo, snapin_tile, txncache_obj, FD_SHMEM_JOIN_MODE_READ_WRITE );
420 0 : for( ulong i=0UL; i<execrp_tile_cnt; i++ ) {
421 0 : fd_topob_tile_uses( topo, &topo->tiles[ fd_topo_find_tile( topo, "execrp", i ) ], txncache_obj, FD_SHMEM_JOIN_MODE_READ_WRITE );
422 0 : }
423 0 : FD_TEST( fd_pod_insertf_ulong( topo->props, txncache_obj->id, "txncache" ) );
424 :
425 0 : if( snapin_tile ) fd_topob_tile_uses( topo, snapin_tile, accdb_obj, FD_SHMEM_JOIN_MODE_READ_WRITE );
426 0 : fd_topob_tile_uses( topo, accdb_tile, accdb_obj, FD_SHMEM_JOIN_MODE_READ_WRITE );
427 0 : fd_topob_tile_uses( topo, replay_tile, accdb_obj, FD_SHMEM_JOIN_MODE_READ_WRITE );
428 0 : FOR( execrp_tile_cnt ) fd_topob_tile_uses( topo, &topo->tiles[ fd_topo_find_tile( topo, "execrp", i ) ], accdb_obj, FD_SHMEM_JOIN_MODE_READ_WRITE );
429 :
430 0 : for( ulong i=0UL; i<topo->tile_cnt; i++ ) {
431 0 : fd_topo_tile_t * tile = &topo->tiles[ i ];
432 0 : fd_topo_configure_tile( tile, config );
433 :
434 0 : if( FD_UNLIKELY( !strcmp( tile->name, "gui" ) ) ) {
435 0 : tile->gui.tile_cnt = topo->tile_cnt;
436 :
437 0 : uchar genesis_hash[ 32 ] = {0};
438 0 : if( FD_LIKELY( -1!=read_genesis_bin( config->paths.genesis, NULL, genesis_hash ) ) ) {
439 0 : char genesis_hash_b58[ FD_BASE58_ENCODED_32_SZ ];
440 0 : fd_base58_encode_32( genesis_hash, NULL, genesis_hash_b58 );
441 0 : strcpy( tile->gui.cluster, fd_genesis_cluster_name( fd_genesis_cluster_identify( genesis_hash_b58 ) ) );
442 0 : }
443 0 : }
444 :
445 0 : if( !strcmp( tile->name, "replay" ) ) {
446 0 : tile->replay.fec_max = config->firedancer.runtime.max_live_slots * fec_sets_per_slot;
447 0 : tile->replay.enable_features_cnt = config->tiles.replay.enable_features_cnt;
448 0 : for( ulong i = 0; i < tile->replay.enable_features_cnt; i++ ) {
449 0 : fd_cstr_ncpy( tile->replay.enable_features[i], config->tiles.replay.enable_features[i], sizeof(tile->replay.enable_features[i]) );
450 0 : }
451 0 : }
452 0 : }
453 :
454 0 : if( FD_LIKELY( telemetry_enabled ) ) wire_event_links( topo );
455 :
456 : // fd_topob_auto_layout( topo, 0 );
457 0 : if( FD_UNLIKELY( !is_auto_affinity ) ) {
458 0 : if( FD_UNLIKELY( affinity_tile_cnt<topo->tile_cnt ) )
459 0 : FD_LOG_ERR(( "topology has %lu tiles but [layout.affinity] only provides %lu entries", topo->tile_cnt, affinity_tile_cnt ));
460 0 : if( FD_UNLIKELY( affinity_tile_cnt>topo->tile_cnt ) )
461 0 : FD_LOG_WARNING(( "topology has %lu tiles but [layout.affinity] provides %lu entries, extra entries will be unused", topo->tile_cnt, affinity_tile_cnt ));
462 0 : }
463 :
464 0 : fd_topob_waker( topo );
465 :
466 0 : fd_topob_finish( topo, CALLBACKS );
467 0 : }
468 :
469 : static void
470 0 : backtest_cmd_topo( config_t * config ) {
471 0 : backtest_topo( config );
472 0 : }
473 :
474 : extern configure_stage_t fd_cfg_stage_keys;
475 :
476 : static args_t
477 0 : configure_args( void ) {
478 0 : args_t args = {
479 0 : .configure.command = CONFIGURE_CMD_INIT,
480 0 : };
481 :
482 : /* FD_BACKTEST_SHARED_HOST skips the host-global stages (irq-affinity,
483 : kworkers) so multiple backtest instances can share a host. */
484 0 : int shared_host = !!getenv( "FD_BACKTEST_SHARED_HOST" );
485 :
486 0 : ulong stage_idx = 0UL;
487 0 : args.configure.stages[ stage_idx++ ] = &fd_cfg_stage_hugetlbfs;
488 0 : if( FD_LIKELY( !shared_host ) ) {
489 0 : args.configure.stages[ stage_idx++ ] = &fd_cfg_stage_irq_affinity;
490 0 : args.configure.stages[ stage_idx++ ] = &fd_cfg_stage_kworkers;
491 0 : }
492 0 : args.configure.stages[ stage_idx++ ] = &fd_cfg_stage_cpuset;
493 0 : args.configure.stages[ stage_idx++ ] = &fd_cfg_stage_snapshots;
494 0 : args.configure.stages[ stage_idx++ ] = &fd_cfg_stage_keys;
495 0 : args.configure.stages[ stage_idx++ ] = NULL;
496 :
497 0 : return args;
498 0 : }
499 :
500 : void
501 : backtest_cmd_perm( args_t * args FD_PARAM_UNUSED,
502 : fd_cap_chk_t * chk,
503 0 : config_t const * config ) {
504 0 : args_t c_args = configure_args();
505 0 : configure_cmd_perm( &c_args, chk, config );
506 0 : run_cmd_perm( NULL, chk, config );
507 0 : }
508 :
509 : static void
510 0 : fixup_config( config_t * config ) {
511 :
512 : /* FIXME Unfortunately, the fdctl boot procedure constructs the
513 : topology before parsing command-line arguments. So, here,
514 : we construct the topology again (a third time ... sigh). */
515 0 : backtest_topo( config );
516 0 : }
517 :
518 : static void
519 : backtest_cmd_fn( args_t * args,
520 0 : config_t * config ) {
521 0 : fixup_config( config );
522 0 : args_t c_args = configure_args();
523 0 : configure_cmd_fn( &c_args, config );
524 :
525 0 : initialize_workspaces( config );
526 0 : initialize_stacks( config );
527 0 : initialize_accdb_fd( config );
528 0 : initialize_store_fds( config );
529 0 : initialize_snapshot_fds( config );
530 :
531 0 : fd_topo_join_workspaces( &config->topo, FD_SHMEM_JOIN_MODE_READ_WRITE, FD_TOPO_CORE_DUMP_LEVEL_DISABLED );
532 0 : fd_topo_fill( &config->topo );
533 :
534 0 : args_t watch_args;
535 0 : int pipefd[2];
536 0 : if( !args->backtest.no_watch ) {
537 0 : if( FD_UNLIKELY( pipe2( pipefd, O_NONBLOCK ) ) ) FD_LOG_ERR(( "pipe2() failed (%i-%s)", errno, fd_io_strerror( errno ) ));
538 :
539 0 : watch_args.watch.drain_output_fd = pipefd[0];
540 0 : watch_args.watch.full = 1;
541 0 : if( FD_UNLIKELY( -1==dup2( pipefd[ 1 ], STDERR_FILENO ) ) ) FD_LOG_ERR(( "dup2() failed (%i-%s)", errno, fd_io_strerror( errno ) ));
542 0 : if( FD_UNLIKELY( -1==close( pipefd[1] ) ) ) FD_LOG_ERR(( "close() failed (%i-%s)", errno, fd_io_strerror( errno ) ));
543 0 : }
544 :
545 0 : fd_topo_run_single_process( &config->topo, 2, config->uid, config->gid, fdctl_tile_run );
546 0 : if( args->backtest.no_watch ) {
547 0 : for(;;) pause();
548 0 : } else {
549 0 : watch_cmd_fn( &watch_args, config );
550 0 : }
551 0 : }
552 :
553 : action_t fd_action_backtest = {
554 : .name = "backtest",
555 : .args = backtest_cmd_args,
556 : .fn = backtest_cmd_fn,
557 : .perm = backtest_cmd_perm,
558 : .topo = backtest_cmd_topo,
559 : .description = "Replay a ledger offline through the replay tile with mocked consensus",
560 : };
|
