Line data Source code
1 : /* The forktest topology exercises offline replay with full consensus.
2 : Constructed using a full topology which is pruned down. */
3 :
4 : #define _GNU_SOURCE
5 : #include "../../../firedancer/topology.h"
6 : #include "../../../shared/fd_action.h"
7 : #include "../../../shared/commands/configure/configure.h"
8 : #include "../../../shared/commands/run/run.h"
9 : #include "../../../shared/commands/watch/watch.h"
10 : #include "../../../../discof/replay/fd_execrp.h"
11 : #include "../../../../discof/genesis/fd_genesi_tile.h"
12 : #include "../../../../discof/tower/fd_tower_tile.h"
13 : #include "../../../../discof/replay/fd_replay_tile.h"
14 : #include "../../../../disco/shred/fd_shred_tile.h"
15 : #include "../../../../disco/net/fd_net_tile.h"
16 : #include "../../../../disco/pack/fd_pack_cost.h"
17 : #include "../../../../disco/topo/fd_topob.h"
18 : #include "../../../../disco/topo/fd_cpu_topo.h"
19 : #include "../../../../util/pod/fd_pod_format.h"
20 : #include "../../../../discof/restore/utils/fd_slot_delta_parser.h"
21 : #include "../../../../discof/restore/utils/fd_ssctrl.h"
22 : #include "../../../../discof/restore/utils/fd_ssmsg.h"
23 : #include "../../../../flamenco/progcache/fd_progcache_admin.h"
24 : #include "../../../../ballet/shred/fd_shred.h"
25 : #include "../../../../discof/backtest/fd_backtest_src.h"
26 : #include "../../../../flamenco/runtime/fd_cost_tracker.h"
27 :
28 : #include <errno.h>
29 : #include <unistd.h> /* pause(2) */
30 : #include <fcntl.h>
31 :
32 : extern fd_topo_obj_callbacks_t * CALLBACKS[];
33 : fd_topo_run_tile_t fdctl_tile_run( fd_topo_tile_t const * tile );
34 :
35 : extern configure_stage_t fd_cfg_stage_keys;
36 :
37 : static args_t
38 0 : configure_args( void ) {
39 0 : args_t args = {
40 0 : .configure.command = CONFIGURE_CMD_INIT,
41 0 : };
42 :
43 0 : ulong stage_idx = 0UL;
44 0 : args.configure.stages[ stage_idx++ ] = &fd_cfg_stage_hugetlbfs;
45 0 : args.configure.stages[ stage_idx++ ] = &fd_cfg_stage_snapshots;
46 0 : args.configure.stages[ stage_idx++ ] = &fd_cfg_stage_keys;
47 0 : args.configure.stages[ stage_idx++ ] = NULL;
48 :
49 0 : return args;
50 0 : }
51 :
52 : static void
53 : forktest_perm( args_t * args FD_PARAM_UNUSED,
54 : fd_cap_chk_t * chk,
55 0 : config_t const * config ) {
56 0 : args_t c_args = configure_args();
57 0 : configure_cmd_perm( &c_args, chk, config );
58 0 : run_cmd_perm( NULL, chk, config );
59 0 : }
60 :
61 : static ushort
62 0 : forktest_recover_expected_shred_version( config_t const * config ) {
63 0 : fd_backtest_src_opts_t opts = {
64 0 : .path = config->firedancer.development.ledger_input.path,
65 0 : .format = config->firedancer.development.ledger_input.format,
66 0 : };
67 :
68 0 : uchar buf[ FD_SHRED_MAX_SZ ];
69 0 : ulong shred_sz = fd_backtest_src_first_shred( &opts, buf, sizeof(buf) );
70 0 : if( FD_UNLIKELY( !shred_sz ) ) FD_LOG_ERR(( "unable to recover shred version from `%s`", opts.path ));
71 :
72 0 : fd_shred_t const * shred = fd_shred_parse( buf, shred_sz, FD_SHRED_BLK_MAX );
73 0 : if( FD_UNLIKELY( !shred ) ) FD_LOG_ERR(( "unable to parse first shred from `%s`", opts.path ));
74 :
75 0 : FD_LOG_INFO(( "Recovered expected shred version %hu", shred->version ));
76 0 : return shred->version;
77 0 : }
78 :
79 : static void
80 0 : forktest_topo( config_t * config ) {
81 :
82 0 : if( FD_UNLIKELY( !config->consensus.expected_shred_version ) ) {
83 0 : config->consensus.expected_shred_version = forktest_recover_expected_shred_version( config );
84 0 : }
85 :
86 0 : config->development.sandbox = 0;
87 0 : config->development.no_clone = 1;
88 :
89 0 : ulong shred_tile_cnt = config->layout.shred_tile_count;
90 :
91 0 : ulong execrp_tile_cnt = config->firedancer.layout.execrp_tile_count;
92 0 : ulong sign_tile_cnt = config->firedancer.layout.sign_tile_count;
93 :
94 0 : int snapshots_enabled = !!config->gossip.entrypoints_cnt;
95 :
96 0 : fd_topo_t * topo = fd_topob_new( &config->topo, config->name );
97 :
98 0 : topo->max_page_size = fd_cstr_to_shmem_page_sz( config->hugetlbfs.max_page_size );
99 0 : topo->gigantic_page_threshold = config->hugetlbfs.gigantic_page_threshold_mib << 20;
100 :
101 : /* topo, name */
102 0 : fd_topob_wksp( topo, "metric" );
103 0 : fd_topob_wksp( topo, "shred" );
104 0 : fd_topob_wksp( topo, "replay" );
105 0 : fd_topob_wksp( topo, "execrp" );
106 0 : fd_topob_wksp( topo, "tower" );
107 0 : fd_topob_wksp( topo, "sign" )->core_dump_level = FD_TOPO_CORE_DUMP_LEVEL_NEVER;
108 :
109 0 : fd_topob_wksp( topo, "metric_in" );
110 :
111 0 : fd_topob_wksp( topo, "net_shred" );
112 :
113 0 : fd_topob_wksp( topo, "gossip_out" );
114 :
115 0 : fd_topob_wksp( topo, "shred_out" );
116 0 : fd_topob_wksp( topo, "replay_epoch" );
117 0 : fd_topob_wksp( topo, "replay_execrp" );
118 0 : fd_topob_wksp( topo, "replay_out" );
119 0 : fd_topob_wksp( topo, "tower_out" );
120 :
121 0 : fd_topob_wksp( topo, "progcache" );
122 0 : fd_topob_wksp( topo, "fec_sets" );
123 0 : fd_topob_wksp( topo, "txncache" );
124 0 : fd_topob_wksp( topo, "banks" );
125 0 : fd_topob_wksp( topo, "store" )->core_dump_level = FD_TOPO_CORE_DUMP_LEVEL_FULL;
126 0 : fd_topob_wksp( topo, "rnonce" );
127 :
128 0 : fd_topob_wksp( topo, "shred_sign" );
129 0 : fd_topob_wksp( topo, "sign_shred" );
130 :
131 0 : if( FD_UNLIKELY( !snapshots_enabled ) ) {
132 0 : fd_topob_wksp( topo, "genesi" );
133 0 : fd_topob_wksp( topo, "genesi_out" );
134 0 : }
135 :
136 0 : fd_topob_wksp( topo, "execrp_replay" );
137 :
138 0 : if( FD_LIKELY( snapshots_enabled ) ) {
139 0 : fd_topob_wksp( topo, "snapct" );
140 0 : fd_topob_wksp( topo, "snapld" );
141 0 : fd_topob_wksp( topo, "snapdc" );
142 0 : fd_topob_wksp( topo, "snapin" );
143 0 : fd_topob_wksp( topo, "snapwr" );
144 0 : fd_topob_wksp( topo, "snapct_ld" );
145 0 : fd_topob_wksp( topo, "snapld_dc" );
146 0 : fd_topob_wksp( topo, "snapdc_in" );
147 :
148 0 : fd_topob_wksp( topo, "snapin_manif" );
149 0 : fd_topob_wksp( topo, "snapct_repr" );
150 :
151 0 : fd_topob_wksp( topo, "snapin_ct" );
152 0 : fd_topob_wksp( topo, "snapwr_ct" );
153 0 : }
154 :
155 0 : fd_topob_wksp( topo, "forkt" );
156 :
157 0 : #define FOR(cnt) for( ulong i=0UL; i<cnt; i++ )
158 :
159 0 : ulong shred_depth = 65536UL; /* from fdctl/topology.c shred_store link. MAKE SURE TO KEEP IN SYNC. */
160 :
161 : /* topo, link_name, wksp_name, depth, mtu, burst */
162 0 : FOR(shred_tile_cnt) fd_topob_link( topo, "shred_net", "net_shred", 32768UL, FD_NET_MTU, 1UL );
163 :
164 0 : if( FD_LIKELY( snapshots_enabled ) ) {
165 : /* TODO: Revisit the depths of all the snapshot links */
166 0 : /**/ fd_topob_link( topo, "snapct_ld", "snapct_ld", 128UL, sizeof(fd_ssctrl_init_t), 1UL );
167 0 : /**/ fd_topob_link( topo, "snapld_dc", "snapld_dc", 16384UL, FD_SNAPSHOT_DATA_MTU, 1UL );
168 0 : /**/ fd_topob_link( topo, "snapdc_in", "snapdc_in", 16384UL, FD_SNAPSHOT_DATA_MTU, 1UL );
169 :
170 : /**/ fd_topob_link( topo, "snapin_manif", "snapin_manif", 4UL, sizeof(fd_snapshot_manifest_t),1UL );
171 0 : /**/ fd_topob_link( topo, "snapct_repr", "snapct_repr", 128UL, 0UL, 1UL )->permit_no_consumers = 1; /* TODO: wire in repair later */
172 :
173 0 : /**/ fd_topob_link( topo, "snapin_ct", "snapin_ct", 128UL, 0UL, 1UL );
174 :
175 : /**/ fd_topob_link( topo, "snapwr_ct", "snapwr_ct", 128UL, 0UL, 1UL );
176 0 : }
177 :
178 0 : if( FD_UNLIKELY( !snapshots_enabled ) ) {
179 0 : /**/ fd_topob_link( topo, "genesi_out", "genesi_out", 1UL, FD_GENESIS_TILE_MTU, 1UL );
180 0 : }
181 :
182 : /**/ fd_topob_link( topo, "gossip_out", "gossip_out", 65536UL*4UL, sizeof(fd_gossip_update_message_t), 1UL ); /* TODO: Unclear where this depth comes from ... fix */
183 :
184 0 : /**/ fd_topob_link( topo, "replay_epoch", "replay_epoch", 16UL, FD_EPOCH_OUT_MTU, 1UL ); /* min pow2 >= replay's STEM_BURST (14); ideally 2, needs per-link burst */
185 0 : /**/ fd_topob_link( topo, "replay_out", "replay_out", 65536UL, sizeof(fd_replay_message_t), 1UL );
186 0 : fd_topob_link( topo, "replay_execrp", "replay_execrp", 16384UL, sizeof(fd_execrp_task_msg_t), 1UL );
187 :
188 0 : FOR(shred_tile_cnt) fd_topob_link( topo, "shred_sign", "shred_sign", 128UL, 32UL, 1UL );
189 0 : FOR(shred_tile_cnt) fd_topob_link( topo, "sign_shred", "sign_shred", 128UL, sizeof(fd_ed25519_sig_t), 1UL );
190 0 : FOR(shred_tile_cnt) fd_topob_link( topo, "shred_out", "shred_out", shred_depth, sizeof(fd_shred_message_t), 3UL ); /* TODO: Pretty sure burst of 3 is incorrect here */
191 0 : /**/ fd_topob_link( topo, "tower_out", "tower_out", 16384UL, sizeof(fd_tower_msg_t), 2UL ); /* conf + slot_done. see explanation in fd_tower_tile.h for link_depth */
192 :
193 0 : FOR(execrp_tile_cnt) fd_topob_link( topo, "execrp_replay", "execrp_replay", 16384UL, sizeof(fd_execrp_task_done_msg_t), 1UL );
194 :
195 0 : ushort parsed_tile_to_cpu[ FD_TILE_MAX ];
196 : /* Unassigned tiles will be floating, unless auto topology is enabled. */
197 0 : for( ulong i=0UL; i<FD_TILE_MAX; i++ ) parsed_tile_to_cpu[ i ] = USHORT_MAX;
198 :
199 0 : int is_auto_affinity = !strcmp( config->firedancer.development.forktest.affinity, "auto" );
200 :
201 0 : fd_topo_cpus_t cpus[1];
202 0 : fd_topo_cpus_init( cpus );
203 :
204 0 : ulong affinity_tile_cnt = 0UL;
205 0 : if( FD_LIKELY( !is_auto_affinity ) ) affinity_tile_cnt = fd_topob_parse_affinity_cstr( config->firedancer.development.forktest.affinity, parsed_tile_to_cpu, 0 );
206 :
207 0 : ulong tile_to_cpu[ FD_TILE_MAX ] = {0};
208 0 : for( ulong i=0UL; i<affinity_tile_cnt; i++ ) {
209 0 : if( FD_UNLIKELY( parsed_tile_to_cpu[ i ]!=USHORT_MAX && parsed_tile_to_cpu[ i ]>=cpus->cpu_cnt ) )
210 0 : FD_LOG_ERR(( "The CPU affinity string in the configuration file under [development.forktest.affinity] specifies a CPU index of %hu, but the system "
211 0 : "only has %lu CPUs. You should either change the CPU allocations in the affinity string, or increase the number of CPUs "
212 0 : "in the system.",
213 0 : parsed_tile_to_cpu[ i ], cpus->cpu_cnt ));
214 0 : tile_to_cpu[ i ] = fd_ulong_if( parsed_tile_to_cpu[ i ]==USHORT_MAX, ULONG_MAX, (ulong)parsed_tile_to_cpu[ i ] );
215 0 : }
216 :
217 0 : fd_topob_link( topo, "net_shred", "net_shred", config->net.ingress_buffer_size, FD_NET_MTU, 1UL );
218 :
219 : /* topo, tile_name, tile_wksp, metrics_wksp, cpu_idx, is_agave, uses_id_keyswitch, uses_av_keyswitch */
220 : /**/ fd_topob_tile( topo, "metric", "metric", "metric_in", tile_to_cpu[ topo->tile_cnt ], 0, 0, 0 );
221 :
222 0 : fd_topo_tile_t * snapin_tile = NULL;
223 0 : if( FD_LIKELY( snapshots_enabled ) ) {
224 0 : /**/ fd_topob_tile( topo, "snapct", "snapct", "metric_in", tile_to_cpu[ topo->tile_cnt ], 0, 0, 0 )->allow_shutdown = 1;
225 0 : /**/ fd_topob_tile( topo, "snapld", "snapld", "metric_in", tile_to_cpu[ topo->tile_cnt ], 0, 0, 0 )->allow_shutdown = 1;
226 0 : /**/ fd_topob_tile( topo, "snapdc", "snapdc", "metric_in", tile_to_cpu[ topo->tile_cnt ], 0, 0, 0 )->allow_shutdown = 1;
227 0 : snapin_tile =
228 0 : /**/ fd_topob_tile( topo, "snapin", "snapin", "metric_in", tile_to_cpu[ topo->tile_cnt ], 0, 0, 0 );
229 0 : snapin_tile->allow_shutdown = 1;
230 0 : /**/ fd_topob_tile( topo, "snapwr", "snapwr", "metric_in", tile_to_cpu[ topo->tile_cnt ], 0, 0, 0 )->allow_shutdown = 1;
231 0 : }
232 :
233 0 : if( FD_UNLIKELY( !snapshots_enabled ) ) {
234 0 : /**/ fd_topob_tile( topo, "genesi", "genesi", "metric_in", tile_to_cpu[ topo->tile_cnt ], 0, 0, 0 )->allow_shutdown = 1;
235 0 : }
236 :
237 0 : FOR(shred_tile_cnt) fd_topob_tile( topo, "shred", "shred", "metric_in", tile_to_cpu[ topo->tile_cnt ], 0, 1, 0 );
238 0 : fd_topo_tile_t * replay_tile =
239 0 : /**/ fd_topob_tile( topo, "replay", "replay", "metric_in", tile_to_cpu[ topo->tile_cnt ], 0, 1, 0 );
240 0 : FOR(execrp_tile_cnt) fd_topob_tile( topo, "execrp", "execrp", "metric_in", tile_to_cpu[ topo->tile_cnt ], 0, 0, 0 );
241 0 : /**/ fd_topob_tile( topo, "tower", "tower", "metric_in", tile_to_cpu[ topo->tile_cnt ], 0, 1, 1 );
242 :
243 0 : FOR(sign_tile_cnt) fd_topob_tile( topo, "sign", "sign", "metric_in", tile_to_cpu[ topo->tile_cnt ], 0, 1, 1 );
244 :
245 0 : fd_topob_tile( topo, "forkt", "forkt", "metric_in", tile_to_cpu[ topo->tile_cnt ], 0, 0, 0 );
246 :
247 0 : fd_topob_wksp( topo, "accdb" )->core_dump_level = FD_TOPO_CORE_DUMP_LEVEL_FULL;
248 0 : fd_topo_tile_t * accdb_tile = fd_topob_tile( topo, "accdb", "accdb", "metric_in", tile_to_cpu[ topo->tile_cnt ], 0, 0, 0 );
249 :
250 0 : fd_topob_wksp( topo, "accdb_data" );
251 0 : fd_topo_obj_t * accdb_obj = setup_topo_accdb( topo, "accdb_data",
252 0 : config->firedancer.accounts.max_accounts,
253 0 : config->firedancer.runtime.max_live_slots,
254 0 : FD_RUNTIME_MAX_WRITABLE_ACCOUNTS_PER_SLOT,
255 0 : 8192UL,
256 0 : 1UL<<35UL,
257 0 : config->firedancer.accounts.cache_size_gib*(1UL<<30UL),
258 0 : config->tiles.bundle.enabled,
259 0 : execrp_tile_cnt+3UL,
260 0 : 0UL );
261 0 : FD_TEST( fd_pod_insertf_ulong( topo->props, accdb_obj->id, "accdb" ) );
262 :
263 0 : if( FD_LIKELY( snapshots_enabled ) ) {
264 0 : fd_topob_tile_out( topo, "snapin", 0UL, "snapin_ct", 0UL );
265 0 : fd_topob_tile_in ( topo, "snapct", 0UL, "metric_in", "snapin_ct", 0UL, FD_TOPOB_RELIABLE, FD_TOPOB_POLLED );
266 :
267 0 : /**/ fd_topob_tile_in ( topo, "snapld", 0UL, "metric_in", "snapct_ld", 0UL, FD_TOPOB_RELIABLE, FD_TOPOB_POLLED );
268 0 : /**/ fd_topob_tile_out( topo, "snapld", 0UL, "snapld_dc", 0UL );
269 :
270 0 : /**/ fd_topob_tile_in ( topo, "snapdc", 0UL, "metric_in", "snapld_dc", 0UL, FD_TOPOB_RELIABLE, FD_TOPOB_POLLED );
271 0 : /**/ fd_topob_tile_out( topo, "snapdc", 0UL, "snapdc_in", 0UL );
272 :
273 0 : fd_topob_tile_in ( topo, "snapin", 0UL, "metric_in", "snapdc_in", 0UL, FD_TOPOB_RELIABLE, FD_TOPOB_POLLED );
274 0 : fd_topob_tile_in ( topo, "snapwr", 0UL, "metric_in", "snapdc_in", 0UL, FD_TOPOB_RELIABLE, FD_TOPOB_POLLED );
275 0 : fd_topob_tile_out( topo, "snapin", 0UL, "snapin_manif", 0UL );
276 :
277 0 : fd_topob_tile_in ( topo, "snapct", 0UL, "metric_in", "snapin_ct", 0UL, FD_TOPOB_RELIABLE, FD_TOPOB_POLLED );
278 0 : fd_topob_tile_in ( topo, "snapct", 0UL, "metric_in", "snapwr_ct", 0UL, FD_TOPOB_RELIABLE, FD_TOPOB_POLLED );
279 0 : fd_topob_tile_in ( topo, "snapct", 0UL, "metric_in", "snapld_dc", 0UL, FD_TOPOB_RELIABLE, FD_TOPOB_POLLED );
280 0 : fd_topob_tile_out( topo, "snapct", 0UL, "snapct_ld", 0UL );
281 0 : fd_topob_tile_out( topo, "snapct", 0UL, "snapct_repr", 0UL );
282 :
283 0 : fd_topob_tile_out( topo, "snapin", 0UL, "snapin_ct", 0UL );
284 0 : fd_topob_tile_out( topo, "snapwr", 0UL, "snapwr_ct", 0UL );
285 0 : }
286 :
287 0 : /**/ fd_topob_tile_in ( topo, "forkt", 0UL, "metric_in", "replay_epoch", 0UL, FD_TOPOB_RELIABLE, FD_TOPOB_POLLED );
288 0 : /**/ fd_topob_tile_out( topo, "forkt", 0UL, "gossip_out", 0UL );
289 0 : /**/ fd_topob_tile_in ( topo, "forkt", 0UL, "metric_in", "tower_out", 0UL, FD_TOPOB_RELIABLE, FD_TOPOB_POLLED );
290 0 : /**/ fd_topob_tile_in ( topo, "forkt", 0UL, "metric_in", "replay_out", 0UL, FD_TOPOB_RELIABLE, FD_TOPOB_POLLED );
291 :
292 0 : FOR(shred_tile_cnt) fd_topob_tile_in ( topo, "shred", i, "metric_in", "net_shred", 0UL, FD_TOPOB_RELIABLE, FD_TOPOB_POLLED );
293 0 : FOR(shred_tile_cnt) fd_topos_tile_in_net( topo, "metric_in", "shred_net", i, FD_TOPOB_RELIABLE, FD_TOPOB_POLLED ); /* No reliable consumers of networking fragments, may be dropped or overrun */
294 :
295 0 : if( FD_UNLIKELY( !snapshots_enabled ) ) {
296 0 : /**/ fd_topob_tile_out( topo, "genesi", 0UL, "genesi_out", 0UL );
297 0 : /**/ fd_topob_tile_in ( topo, "replay", 0UL, "metric_in", "genesi_out", 0UL, FD_TOPOB_RELIABLE, FD_TOPOB_POLLED );
298 0 : }
299 0 : /**/ fd_topob_tile_out( topo, "replay", 0UL, "replay_out", 0UL );
300 0 : /**/ fd_topob_tile_out( topo, "replay", 0UL, "replay_epoch", 0UL );
301 0 : /**/ fd_topob_tile_out( topo, "replay", 0UL, "replay_execrp", 0UL );
302 0 : FOR(execrp_tile_cnt) fd_topob_tile_in ( topo, "replay", 0UL, "metric_in", "execrp_replay", i, FD_TOPOB_RELIABLE, FD_TOPOB_POLLED );
303 0 : /**/ fd_topob_tile_in ( topo, "replay", 0UL, "metric_in", "tower_out", 0UL, FD_TOPOB_RELIABLE, FD_TOPOB_POLLED );
304 0 : /**/ fd_topob_tile_in ( topo, "replay", 0UL, "metric_in", "gossip_out", 0UL, FD_TOPOB_RELIABLE, FD_TOPOB_POLLED );
305 0 : if( FD_LIKELY( snapshots_enabled ) ) {
306 0 : fd_topob_tile_in ( topo, "replay", 0UL, "metric_in", "snapin_manif", 0UL, FD_TOPOB_RELIABLE, FD_TOPOB_POLLED );
307 0 : }
308 :
309 0 : FOR(execrp_tile_cnt) fd_topob_tile_in ( topo, "execrp", i, "metric_in", "replay_execrp", 0UL, FD_TOPOB_RELIABLE, FD_TOPOB_POLLED );
310 0 : FOR(execrp_tile_cnt) fd_topob_tile_out( topo, "execrp", i, "execrp_replay", i );
311 :
312 0 : /**/ fd_topob_tile_in ( topo, "tower", 0UL, "metric_in", "replay_epoch", 0UL, FD_TOPOB_RELIABLE, FD_TOPOB_POLLED );
313 0 : /**/ fd_topob_tile_in ( topo, "tower", 0UL, "metric_in", "gossip_out", 0UL, FD_TOPOB_RELIABLE, FD_TOPOB_POLLED );
314 0 : /**/ fd_topob_tile_in ( topo, "tower", 0UL, "metric_in", "replay_out", 0UL, FD_TOPOB_RELIABLE, FD_TOPOB_POLLED );
315 0 : FOR(shred_tile_cnt) fd_topob_tile_in( topo, "tower", 0UL, "metric_in", "shred_out", i, FD_TOPOB_RELIABLE, FD_TOPOB_POLLED );
316 0 : FOR(shred_tile_cnt) fd_topob_tile_in( topo, "replay", 0UL, "metric_in", "shred_out", i, FD_TOPOB_RELIABLE, FD_TOPOB_POLLED );
317 0 : /**/ fd_topob_tile_out( topo, "tower", 0UL, "tower_out", 0UL );
318 :
319 0 : FOR(shred_tile_cnt) fd_topob_tile_in ( topo, "shred", i, "metric_in", "replay_epoch", 0UL, FD_TOPOB_RELIABLE, FD_TOPOB_POLLED );
320 0 : FOR(shred_tile_cnt) fd_topob_tile_in ( topo, "shred", i, "metric_in", "gossip_out", 0UL, FD_TOPOB_RELIABLE, FD_TOPOB_POLLED );
321 0 : FOR(shred_tile_cnt) fd_topob_tile_out( topo, "shred", i, "shred_out", i );
322 0 : FOR(shred_tile_cnt) fd_topob_tile_in ( topo, "shred", i, "metric_in", "tower_out", 0UL, FD_TOPOB_RELIABLE, FD_TOPOB_POLLED );
323 0 : FOR(shred_tile_cnt) fd_topob_tile_out( topo, "shred", i, "shred_net", i );
324 :
325 : /* topo, tile_name, tile_kind_id, fseq_wksp, link_name, link_kind_id, reliable, polled */
326 :
327 0 : for( ulong i=0UL; i<shred_tile_cnt; i++ ) {
328 0 : /**/ fd_topob_tile_in ( topo, "sign", 0UL, "metric_in", "shred_sign", i, FD_TOPOB_UNRELIABLE, FD_TOPOB_POLLED );
329 0 : /**/ fd_topob_tile_out( topo, "shred", i, "shred_sign", i );
330 0 : /**/ fd_topob_tile_in ( topo, "shred", i, "metric_in", "sign_shred", i, FD_TOPOB_UNRELIABLE, FD_TOPOB_UNPOLLED );
331 0 : /**/ fd_topob_tile_out( topo, "sign", 0UL, "sign_shred", i );
332 0 : }
333 :
334 0 : fd_topob_tile_out( topo, "forkt", 0UL, "net_shred", 0UL );
335 0 : FOR(shred_tile_cnt) fd_topob_tile_in( topo, "forkt", 0UL, "metric_in", "shred_net", 0UL, FD_TOPOB_RELIABLE, FD_TOPOB_POLLED );
336 :
337 0 : if( FD_LIKELY( !is_auto_affinity ) ) {
338 0 : if( FD_UNLIKELY( affinity_tile_cnt<topo->tile_cnt ) )
339 0 : FD_LOG_ERR(( "The topology you are using has %lu tiles, but the CPU affinity specified in the config tile as [development.forktest.affinity] only provides for %lu cores. "
340 0 : "You should either increase the number of cores dedicated to Firedancer in the affinity string, or decrease the number of cores needed by reducing "
341 0 : "the total tile count. You can reduce the tile count by decreasing individual tile counts in the [layout] section of the configuration file.",
342 0 : topo->tile_cnt, affinity_tile_cnt ));
343 0 : if( FD_UNLIKELY( affinity_tile_cnt>topo->tile_cnt ) )
344 0 : FD_LOG_WARNING(( "The topology you are using has %lu tiles, but the CPU affinity specified in the config tile as [development.forktest.affinity] provides for %lu cores. "
345 0 : "Not all cores in the affinity will be used by Firedancer. You may wish to increase the number of tiles in the system by increasing "
346 0 : "individual tile counts in the [layout] section of the configuration file.",
347 0 : topo->tile_cnt, affinity_tile_cnt ));
348 0 : } else {
349 0 : ushort blocklist_cores[ FD_TILE_MAX ];
350 0 : topo->blocklist_cores_cnt = fd_topob_parse_affinity_cstr( config->layout.blocklist_cores, blocklist_cores, 0 );
351 0 : if( FD_UNLIKELY( topo->blocklist_cores_cnt>FD_TILE_MAX ) ) {
352 0 : FD_LOG_ERR(( "The CPU string in the configuration file under [layout.blocklist_cores] specifies more CPUs than Firedancer can use. "
353 0 : "You should reduce the number of CPUs in the excluded cores string." ));
354 0 : }
355 :
356 0 : for( ulong i=0UL; i<topo->blocklist_cores_cnt; i++ ) {
357 0 : if( FD_UNLIKELY( blocklist_cores[ i ]==USHORT_MAX ) ) {
358 0 : FD_LOG_ERR(( "The CPU string in the configuration file under [layout.blocklist_cores] contains invalid values: `f`. "
359 0 : "You should fix the excluded cores string." ));
360 0 : }
361 0 : topo->blocklist_cores_cpu_idx[ i ] = blocklist_cores[ i ];
362 0 : }
363 0 : }
364 :
365 0 : if( FD_UNLIKELY( is_auto_affinity ) ) fd_topob_auto_layout( topo, 0 );
366 :
367 : /* Repair and shred share a secret they use to generate the nonces. */
368 0 : if( 1 ) {
369 0 : fd_topo_obj_t * rnonce_ss_obj = fd_topob_obj( topo, "rnonce_ss", "rnonce" );
370 0 : for( ulong i=0UL; i<shred_tile_cnt; i++ ) {
371 0 : fd_topo_tile_t * shred_tile = &topo->tiles[ fd_topo_find_tile( topo, "shred", i ) ];
372 0 : fd_topob_tile_uses( topo, shred_tile, rnonce_ss_obj, FD_SHMEM_JOIN_MODE_READ_ONLY );
373 0 : }
374 0 : FD_TEST( fd_pod_insertf_ulong( topo->props, rnonce_ss_obj->id, "rnonce_ss" ) );
375 0 : }
376 :
377 0 : setup_topo_progcache( topo, "progcache",
378 0 : fd_progcache_est_rec_max( config->firedancer.runtime.program_cache.heap_size_mib<<20,
379 0 : config->firedancer.runtime.program_cache.mean_cache_entry_size ),
380 0 : config->firedancer.runtime.max_live_slots,
381 0 : config->firedancer.runtime.program_cache.heap_size_mib<<20 );
382 0 : ulong progcache_obj_id; FD_TEST( (progcache_obj_id = fd_pod_query_ulong( topo->props, "progcache", ULONG_MAX ))!=ULONG_MAX );
383 0 : fd_topo_obj_t * progcache_obj = &topo->objs[ progcache_obj_id ];
384 :
385 0 : /**/ fd_topob_tile_uses( topo, replay_tile, progcache_obj, FD_SHMEM_JOIN_MODE_READ_WRITE );
386 0 : FOR(execrp_tile_cnt) fd_topob_tile_uses( topo, &topo->tiles[ fd_topo_find_tile( topo, "execrp", i ) ], progcache_obj, FD_SHMEM_JOIN_MODE_READ_WRITE );
387 :
388 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.larger_max_cost_per_block );
389 0 : /**/ fd_topob_tile_uses( topo, replay_tile, banks_obj, FD_SHMEM_JOIN_MODE_READ_WRITE );
390 0 : /**/ fd_topob_tile_uses( topo, &topo->tiles[ fd_topo_find_tile( topo, "tower", 0UL ) ], banks_obj, FD_SHMEM_JOIN_MODE_READ_ONLY );
391 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 );
392 0 : FD_TEST( fd_pod_insertf_ulong( topo->props, banks_obj->id, "banks" ) );
393 :
394 0 : fd_topo_obj_t * node_info_obj = fd_topob_obj( topo, "node_info", "replay" );
395 0 : fd_topob_tile_uses( topo, replay_tile, node_info_obj, FD_SHMEM_JOIN_MODE_READ_WRITE );
396 0 : fd_topob_tile_uses( topo, &topo->tiles[ fd_topo_find_tile( topo, "tower", 0UL ) ], node_info_obj, FD_SHMEM_JOIN_MODE_READ_WRITE );
397 0 : FD_TEST( fd_pod_insertf_ulong( topo->props, node_info_obj->id, "node_info" ) );
398 :
399 0 : ulong fec_set_cnt = 2UL*shred_depth + config->tiles.shred.max_pending_shred_sets + 6UL;
400 0 : ulong fec_sets_sz = fec_set_cnt*sizeof(fd_fec_set_t); /* mirrors # of dcache entries in frankendancer */
401 0 : fd_topo_obj_t * fec_sets_obj = setup_topo_fec_sets( topo, "fec_sets", shred_tile_cnt*fec_sets_sz );
402 0 : for( ulong i=0UL; i<shred_tile_cnt; i++ ) {
403 0 : fd_topo_tile_t * shred_tile = &topo->tiles[ fd_topo_find_tile( topo, "shred", i ) ];
404 0 : fd_topob_tile_uses( topo, shred_tile, fec_sets_obj, FD_SHMEM_JOIN_MODE_READ_WRITE );
405 0 : }
406 0 : FD_TEST( fd_pod_insertf_ulong( topo->props, fec_sets_obj->id, "fec_sets" ) );
407 :
408 0 : ulong store_fec_max = config->firedancer.runtime.max_live_slots * FD_FEC_BLK_MAX + (shred_depth * shred_tile_cnt) + 1;
409 0 : ulong store_fec_data_max = fd_ulong_if( config->firedancer.development.fixed_fec_sets, 31840UL, 63985UL );
410 0 : fd_topo_obj_t * store_obj = setup_topo_store( topo, "store", store_fec_max, (uint)shred_tile_cnt, store_fec_data_max );
411 0 : FOR(shred_tile_cnt) fd_topob_tile_uses( topo, &topo->tiles[ fd_topo_find_tile( topo, "shred", i ) ], store_obj, FD_SHMEM_JOIN_MODE_READ_WRITE );
412 0 : fd_topob_tile_uses( topo, replay_tile, store_obj, FD_SHMEM_JOIN_MODE_READ_WRITE );
413 0 : FD_TEST( fd_pod_insertf_ulong( topo->props, store_obj->id, "store" ) );
414 :
415 0 : fd_topo_obj_t * txncache_obj = setup_topo_txncache( topo, "txncache", config->firedancer.runtime.max_live_slots, FD_PACK_MAX_TXNCACHE_TXN_PER_SLOT, config->development.bench.larger_max_cost_per_block );
416 0 : fd_topob_tile_uses( topo, replay_tile, txncache_obj, FD_SHMEM_JOIN_MODE_READ_WRITE );
417 0 : if( FD_LIKELY( snapshots_enabled ) ) {
418 0 : fd_topob_tile_uses( topo, snapin_tile, txncache_obj, FD_SHMEM_JOIN_MODE_READ_WRITE );
419 0 : }
420 0 : FOR(execrp_tile_cnt) fd_topob_tile_uses( topo, &topo->tiles[ fd_topo_find_tile( topo, "execrp", i ) ], txncache_obj, FD_SHMEM_JOIN_MODE_READ_WRITE );
421 0 : FD_TEST( fd_pod_insertf_ulong( topo->props, txncache_obj->id, "txncache" ) );
422 :
423 0 : if( FD_LIKELY( snapshots_enabled ) ) fd_topob_tile_uses( topo, snapin_tile, accdb_obj, FD_SHMEM_JOIN_MODE_READ_WRITE );
424 0 : fd_topob_tile_uses( topo, accdb_tile, accdb_obj, FD_SHMEM_JOIN_MODE_READ_WRITE );
425 0 : fd_topob_tile_uses( topo, replay_tile, accdb_obj, FD_SHMEM_JOIN_MODE_READ_WRITE );
426 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 );
427 :
428 0 : fd_pod_insert_int( topo->props, "sandbox", config->development.sandbox ? 1 : 0 );
429 :
430 0 : for( ulong i=0UL; i<topo->tile_cnt; i++ ) {
431 0 : fd_topo_configure_tile( &topo->tiles[ i ], config );
432 0 : }
433 :
434 0 : fd_topob_finish( topo, CALLBACKS );
435 0 : }
436 :
437 : static void
438 : forktest_cmd_args( int * pargc,
439 : char *** pargv,
440 0 : args_t * args ) {
441 0 : args->forktest.no_watch = fd_env_strip_cmdline_contains( pargc, pargv, "--no-watch" );
442 0 : }
443 :
444 : static void
445 : forktest_fn( args_t * args,
446 0 : config_t * config ) {
447 0 : args_t c_args = configure_args();
448 0 : configure_cmd_fn( &c_args, config );
449 :
450 0 : initialize_workspaces( config );
451 0 : initialize_stacks( config );
452 0 : initialize_accdb_fd( config );
453 0 : initialize_snapshot_fds( config );
454 :
455 0 : fd_topo_join_workspaces( &config->topo, FD_SHMEM_JOIN_MODE_READ_WRITE, FD_TOPO_CORE_DUMP_LEVEL_DISABLED );
456 0 : fd_topo_fill( &config->topo );
457 :
458 0 : args_t watch_args;
459 0 : int pipefd[2];
460 0 : if( !args->forktest.no_watch ) {
461 0 : if( FD_UNLIKELY( pipe2( pipefd, O_NONBLOCK ) ) ) FD_LOG_ERR(( "pipe2() failed (%i-%s)", errno, fd_io_strerror( errno ) ));
462 :
463 0 : watch_args.watch.drain_output_fd = pipefd[0];
464 0 : watch_args.watch.full = 1;
465 0 : if( FD_UNLIKELY( -1==dup2( pipefd[1], STDERR_FILENO ) ) ) FD_LOG_ERR(( "dup2() failed (%i-%s)", errno, fd_io_strerror( errno ) ));
466 0 : if( FD_UNLIKELY( -1==close( pipefd[1] ) ) ) FD_LOG_ERR(( "close() failed (%i-%s)", errno, fd_io_strerror( errno ) ));
467 0 : }
468 :
469 0 : fd_topo_run_single_process( &config->topo, 2, config->uid, config->gid, fdctl_tile_run );
470 0 : if( args->forktest.no_watch ) {
471 0 : for(;;) pause();
472 0 : } else {
473 0 : watch_cmd_fn( &watch_args, config );
474 0 : }
475 0 : }
476 :
477 : action_t fd_action_forktest = {
478 : .name = "forktest",
479 : .args = forktest_cmd_args,
480 : .fn = forktest_fn,
481 : .perm = forktest_perm,
482 : .topo = forktest_topo,
483 : .is_local_cluster = 1,
484 : .description = "Replay a ledger offline through a pruned full topology with real consensus",
485 : };
|
