Line data Source code
1 : /* The repair command spawns a smaller topology for profiling the repair
2 : tile. This is a standalone application, and it can be run in mainnet,
3 : testnet and/or a private cluster. */
4 :
5 : #include "../../../disco/net/fd_net_tile.h"
6 : #include "../../../disco/tiles.h"
7 : #include "../../../disco/topo/fd_topob.h"
8 : #include "../../../disco/topo/fd_cpu_topo.h"
9 : #include "../../../util/pod/fd_pod_format.h"
10 :
11 : #include "../../firedancer/topology.h"
12 : #include "../../shared/commands/configure/configure.h"
13 : #include "../../shared/commands/run/run.h" /* initialize_workspaces */
14 : #include "../../shared/fd_config.h" /* config_t */
15 : #include "../../shared_dev/commands/dev.h"
16 : #include "../../../disco/tiles.h"
17 : #include "../../../disco/topo/fd_topob.h"
18 : #include "../../../util/pod/fd_pod_format.h"
19 : #include "../../../waltz/resolv/fd_io_readline.h"
20 : #include "../../platform/fd_sys_util.h"
21 : #include "../../shared/commands/monitor/helper.h"
22 : #include "../../../disco/metrics/fd_metrics.h"
23 : #include "../../../discof/restore/utils/fd_ssmanifest_parser.h"
24 : #include "../../../flamenco/runtime/sysvar/fd_sysvar_epoch_schedule.h"
25 : #include "../../../flamenco/stakes/fd_stake_weight.h"
26 : #include "../../../flamenco/leaders/fd_leaders_base.h"
27 : #include "../../../discof/repair/fd_repair_tile.c"
28 :
29 : #include "gossip.h"
30 : #include "core_subtopo.h"
31 :
32 : #include <unistd.h> /* pause */
33 : #include <fcntl.h>
34 : #include <stdio.h>
35 : #include <termios.h>
36 : #include <errno.h>
37 :
38 : extern action_t fd_action_repair;
39 :
40 : struct fd_location_info {
41 : ulong ip4_addr; /* for map key convenience */
42 : char location[ 128 ];
43 : };
44 : typedef struct fd_location_info fd_location_info_t;
45 :
46 : #define MAP_NAME fd_location_table
47 0 : #define MAP_T fd_location_info_t
48 0 : #define MAP_KEY ip4_addr
49 0 : #define MAP_LG_SLOT_CNT 16
50 : #define MAP_MEMOIZE 0
51 : #include "../../../util/tmpl/fd_map.c"
52 :
53 : uchar __attribute__((aligned(alignof(fd_location_info_t)))) location_table_mem[ sizeof(fd_location_info_t) * (1 << 16 ) ];
54 :
55 : static struct termios termios_backup;
56 :
57 : static void
58 0 : restore_terminal( void ) {
59 0 : (void)tcsetattr( STDIN_FILENO, TCSANOW, &termios_backup );
60 0 : }
61 :
62 : extern fd_topo_obj_callbacks_t * CALLBACKS[];
63 :
64 : fd_topo_run_tile_t
65 : fdctl_tile_run( fd_topo_tile_t const * tile );
66 :
67 0 : #define MANIFEST_LOAD_MAX_SZ (2UL * FD_SHMEM_GIGANTIC_PAGE_SZ)
68 :
69 : /* https://github.com/anza-xyz/agave/blob/v3.1.8/runtime/src/snapshot_bank_utils.rs#L632 */
70 : static int
71 0 : repair_verify_epoch_stakes( fd_snapshot_manifest_t const * manifest ) {
72 0 : fd_epoch_schedule_t epoch_schedule = (fd_epoch_schedule_t){
73 0 : .slots_per_epoch = manifest->epoch_schedule_params.slots_per_epoch,
74 0 : .leader_schedule_slot_offset = manifest->epoch_schedule_params.leader_schedule_slot_offset,
75 0 : .warmup = manifest->epoch_schedule_params.warmup,
76 0 : .first_normal_epoch = manifest->epoch_schedule_params.first_normal_epoch,
77 0 : .first_normal_slot = manifest->epoch_schedule_params.first_normal_slot,
78 0 : };
79 :
80 0 : ulong min_required_epoch = fd_slot_to_epoch( &epoch_schedule, manifest->slot, NULL );
81 0 : ulong max_required_epoch = fd_slot_to_leader_schedule_epoch( &epoch_schedule, manifest->slot );
82 :
83 0 : for( ulong i=min_required_epoch; i<=max_required_epoch; i++ ) {
84 0 : int found = 0;
85 0 : for( ulong j=0UL; j<FD_RUNTIME_MANIFEST_EPOCH_STAKES_LEN; j++ ) {
86 0 : if( manifest->epoch_stakes[j].epoch==i ) {
87 0 : found = 1;
88 0 : break;
89 0 : }
90 0 : }
91 0 : if( FD_UNLIKELY( !found ) ) {
92 0 : FD_LOG_WARNING(( "stakes not found for epoch %lu in manifest", i ));
93 0 : return -1;
94 0 : }
95 0 : }
96 0 : return 0;
97 0 : }
98 :
99 : static inline ulong
100 : repair_generate_epoch_info_msg( ulong epoch,
101 : fd_epoch_schedule_t const * epoch_schedule,
102 : fd_snapshot_manifest_epoch_stakes_t const * epoch_stakes,
103 0 : ulong * epoch_info_msg_out ) {
104 0 : fd_epoch_info_msg_t * epoch_info_msg = (fd_epoch_info_msg_t *)fd_type_pun( epoch_info_msg_out );
105 0 : fd_vote_stake_weight_t * stake_weights = fd_epoch_info_msg_stake_weights( epoch_info_msg );
106 :
107 0 : epoch_info_msg->epoch = epoch;
108 0 : epoch_info_msg->start_slot = fd_epoch_slot0( epoch_schedule, epoch );
109 0 : epoch_info_msg->slot_cnt = fd_epoch_slot_cnt( epoch_schedule, epoch );
110 0 : epoch_info_msg->ns_per_slot = FD_SLOT_PARAMS_400MS.ns_per_slot;
111 :
112 0 : fd_memset( &epoch_info_msg->features, 0xFF, sizeof(fd_features_t) );
113 :
114 0 : ulong idx = 0UL;
115 0 : for( ulong i=0UL; i<epoch_stakes->vote_stakes_len; i++ ) {
116 0 : ulong stake = epoch_stakes->vote_stakes[ i ].stake;
117 0 : if( FD_UNLIKELY( !stake ) ) continue;
118 0 : stake_weights[ idx ].stake = stake;
119 0 : memcpy( stake_weights[ idx ].id_key.uc, epoch_stakes->vote_stakes[ i ].identity, sizeof(fd_pubkey_t) );
120 0 : memcpy( stake_weights[ idx ].vote_key.uc, epoch_stakes->vote_stakes[ i ].vote, sizeof(fd_pubkey_t) );
121 0 : idx++;
122 0 : }
123 0 : epoch_info_msg->staked_vote_cnt = idx;
124 0 : sort_vote_weights_by_stake_vote_inplace( stake_weights, idx );
125 :
126 0 : fd_stake_weight_t * id_weights = fd_epoch_info_msg_id_weights( epoch_info_msg );
127 0 : epoch_info_msg->staked_id_cnt = compute_id_weights_from_vote_weights( id_weights, stake_weights, epoch_info_msg->staked_vote_cnt );
128 0 : FD_TEST( idx<=MAX_SHRED_DESTS );
129 :
130 0 : epoch_info_msg->epoch_schedule = *epoch_schedule;
131 0 : return fd_epoch_info_msg_sz( epoch_info_msg->staked_vote_cnt, epoch_info_msg->staked_id_cnt );
132 0 : }
133 :
134 : /* repair_load_manifest loads the snapshot manifest from disk and
135 : pre-populates the snapin_manif and replay_epoch dcache links so
136 : that consumer tiles see the data on their first poll cycle. */
137 : static void
138 : repair_load_manifest( fd_topo_t * topo,
139 0 : char const * manifest_path ) {
140 0 : if( FD_UNLIKELY( !manifest_path || !manifest_path[0] ) ) return;
141 :
142 : /* Parse manifest */
143 :
144 0 : int fd = open( manifest_path, O_RDONLY );
145 0 : if( FD_UNLIKELY( fd<0 ) ) FD_LOG_ERR(( "open(%s) failed (%d-%s)", manifest_path, errno, fd_io_strerror( errno ) ));
146 :
147 0 : fd_snapshot_manifest_t * manifest = aligned_alloc( alignof(fd_snapshot_manifest_t), sizeof(fd_snapshot_manifest_t) );
148 0 : FD_TEST( manifest );
149 0 : for( ulong i=0UL; i<FD_RUNTIME_MANIFEST_EPOCH_STAKES_LEN; i++ ) manifest->epoch_stakes[i].epoch = ULONG_MAX;
150 :
151 0 : uchar * buf = aligned_alloc( 128UL, MANIFEST_LOAD_MAX_SZ );
152 0 : FD_TEST( buf );
153 0 : ulong buf_sz = 0;
154 0 : FD_TEST( !fd_io_read( fd, buf, 0UL, MANIFEST_LOAD_MAX_SZ-1UL, &buf_sz ) );
155 0 : close( fd );
156 :
157 0 : fd_ssmanifest_parser_t * parser = fd_ssmanifest_parser_join( fd_ssmanifest_parser_new(
158 0 : aligned_alloc( fd_ssmanifest_parser_align(), fd_ssmanifest_parser_footprint() ) ) );
159 0 : FD_TEST( parser );
160 0 : fd_ssmanifest_parser_init( parser, manifest );
161 0 : int parser_err = fd_ssmanifest_parser_consume( parser, buf, buf_sz );
162 0 : FD_TEST( parser_err!=FD_SSMANIFEST_PARSER_ADVANCE_ERROR );
163 0 : FD_TEST( fd_ssmanifest_parser_fini( parser )==FD_SSMANIFEST_PARSER_ADVANCE_DONE );
164 0 : free( parser );
165 0 : free( buf );
166 :
167 0 : FD_LOG_NOTICE(( "manifest bank slot %lu", manifest->slot ));
168 0 : FD_TEST( !repair_verify_epoch_stakes( manifest ) );
169 :
170 : /* Update root_slot fseq */
171 :
172 0 : ulong root_slot_obj_id = fd_pod_queryf_ulong( topo->props, ULONG_MAX, "root_slot" );
173 0 : if( FD_LIKELY( root_slot_obj_id!=ULONG_MAX ) ) {
174 0 : ulong * root_fseq = fd_fseq_join( fd_topo_obj_laddr( topo, root_slot_obj_id ) );
175 0 : FD_TEST( root_fseq );
176 0 : fd_fseq_update( root_fseq, manifest->slot );
177 0 : }
178 :
179 : /* Publish manifest to snapin_manif dcache */
180 :
181 0 : ulong snap_link_idx = fd_topo_find_link( topo, "snapin_manif", 0UL );
182 0 : FD_TEST( snap_link_idx!=ULONG_MAX );
183 0 : fd_topo_link_t * snap_link = &topo->links[ snap_link_idx ];
184 0 : fd_wksp_t * snap_mem = topo->workspaces[ topo->objs[ snap_link->dcache_obj_id ].wksp_id ].wksp;
185 0 : ulong snap_chunk0 = fd_dcache_compact_chunk0( snap_mem, snap_link->dcache );
186 0 : ulong snap_wmark = fd_dcache_compact_wmark ( snap_mem, snap_link->dcache, snap_link->mtu );
187 0 : ulong snap_chunk = snap_chunk0;
188 :
189 0 : uchar * snap_dst = fd_chunk_to_laddr( snap_mem, snap_chunk );
190 0 : memcpy( snap_dst, manifest, sizeof(fd_snapshot_manifest_t) );
191 0 : fd_mcache_publish( snap_link->mcache, snap_link->depth, 0UL,
192 0 : fd_ssmsg_sig( FD_SSMSG_MANIFEST_INCREMENTAL ),
193 0 : snap_chunk, sizeof(fd_snapshot_manifest_t), 0UL, 0UL, 0UL );
194 0 : snap_chunk = fd_dcache_compact_next( snap_chunk, sizeof(fd_snapshot_manifest_t), snap_chunk0, snap_wmark );
195 :
196 0 : fd_mcache_publish( snap_link->mcache, snap_link->depth, 1UL,
197 0 : fd_ssmsg_sig( FD_SSMSG_DONE ), 0UL, 0UL, 0UL, 0UL, 0UL );
198 :
199 : /* Publish epoch stake weights to replay_epoch dcache */
200 :
201 0 : ulong epoch_link_idx = fd_topo_find_link( topo, "replay_epoch", 0UL );
202 0 : FD_TEST( epoch_link_idx!=ULONG_MAX );
203 0 : fd_topo_link_t * epoch_link = &topo->links[ epoch_link_idx ];
204 0 : fd_wksp_t * epoch_mem = topo->workspaces[ topo->objs[ epoch_link->dcache_obj_id ].wksp_id ].wksp;
205 0 : ulong epoch_chunk0 = fd_dcache_compact_chunk0( epoch_mem, epoch_link->dcache );
206 0 : ulong epoch_wmark = fd_dcache_compact_wmark ( epoch_mem, epoch_link->dcache, epoch_link->mtu );
207 0 : ulong epoch_chunk = epoch_chunk0;
208 0 : ulong epoch_seq = 0UL;
209 :
210 : /* Construct fd_epoch_schedule_t field-by-field rather than type-punning
211 : from the unpacked manifest struct (fd_epoch_schedule_t is packed). */
212 0 : fd_epoch_schedule_t schedule_local;
213 0 : schedule_local.slots_per_epoch = manifest->epoch_schedule_params.slots_per_epoch;
214 0 : schedule_local.leader_schedule_slot_offset = manifest->epoch_schedule_params.leader_schedule_slot_offset;
215 0 : schedule_local.warmup = manifest->epoch_schedule_params.warmup;
216 0 : schedule_local.first_normal_epoch = manifest->epoch_schedule_params.first_normal_epoch;
217 0 : schedule_local.first_normal_slot = manifest->epoch_schedule_params.first_normal_slot;
218 0 : fd_epoch_schedule_t const * schedule = &schedule_local;
219 0 : ulong epoch = fd_slot_to_epoch( schedule, manifest->slot, NULL );
220 :
221 0 : ulong epoch_stakes_base = epoch > 0UL ? epoch - 1UL : 0UL;
222 0 : ulong leader_schedule_epoch = fd_slot_to_leader_schedule_epoch( schedule, manifest->slot );
223 0 : ulong cur_idx = epoch - epoch_stakes_base;
224 0 : FD_TEST( cur_idx < FD_RUNTIME_MANIFEST_EPOCH_STAKES_LEN );
225 :
226 0 : ulong * epoch_dst = fd_chunk_to_laddr( epoch_mem, epoch_chunk );
227 0 : ulong epoch_sz = repair_generate_epoch_info_msg( epoch, schedule, &manifest->epoch_stakes[cur_idx], epoch_dst );
228 0 : fd_mcache_publish( epoch_link->mcache, epoch_link->depth, epoch_seq,
229 0 : 4UL, epoch_chunk, epoch_sz, 0UL, 0UL, fd_frag_meta_ts_comp( fd_tickcount() ) );
230 0 : epoch_chunk = fd_dcache_compact_next( epoch_chunk, epoch_sz, epoch_chunk0, epoch_wmark );
231 0 : epoch_seq++;
232 0 : FD_LOG_NOTICE(( "sending current epoch stake weights - epoch: %lu", epoch ));
233 :
234 0 : if( leader_schedule_epoch >= epoch + 1UL ) {
235 0 : ulong next_idx = epoch + 1UL - epoch_stakes_base;
236 0 : FD_TEST( next_idx < FD_RUNTIME_MANIFEST_EPOCH_STAKES_LEN );
237 :
238 0 : epoch_dst = fd_chunk_to_laddr( epoch_mem, epoch_chunk );
239 0 : epoch_sz = repair_generate_epoch_info_msg( epoch + 1UL, schedule, &manifest->epoch_stakes[next_idx], epoch_dst );
240 0 : fd_mcache_publish( epoch_link->mcache, epoch_link->depth, epoch_seq,
241 0 : 4UL, epoch_chunk, epoch_sz, 0UL, 0UL, fd_frag_meta_ts_comp( fd_tickcount() ) );
242 0 : epoch_chunk = fd_dcache_compact_next( epoch_chunk, epoch_sz, epoch_chunk0, epoch_wmark );
243 0 : epoch_seq++;
244 0 : FD_LOG_NOTICE(( "sending next epoch stake weights - epoch: %lu", epoch + 1UL ));
245 0 : }
246 0 : (void)epoch_chunk;
247 :
248 0 : free( manifest );
249 0 : }
250 :
251 : /* repair_topo is a subset of "src/app/firedancer/topology.c" at commit
252 : 0d8386f4f305bb15329813cfe4a40c3594249e96, slightly modified to work
253 : as a repair catchup. TODO ideally, one should invoke the firedancer
254 : topology first, and exclude the parts that are not needed, instead of
255 : manually generating new topologies for every command. This would
256 : also guarantee that the catchup is replicating (as close as possible)
257 : the full topology. */
258 : static void
259 0 : repair_topo( config_t * config ) {
260 0 : ulong net_tile_cnt = config->layout.net_tile_count;
261 0 : ulong shred_tile_cnt = config->layout.shred_tile_count;
262 0 : ulong quic_tile_cnt = config->layout.quic_tile_count;
263 0 : ulong sign_tile_cnt = config->firedancer.layout.sign_tile_count;
264 0 : ulong gossvf_tile_cnt = config->firedancer.layout.gossvf_tile_count;
265 :
266 0 : fd_topo_t * topo = { fd_topob_new( &config->topo, config->name ) };
267 0 : topo->max_page_size = fd_cstr_to_shmem_page_sz( config->hugetlbfs.max_page_size );
268 0 : topo->gigantic_page_threshold = config->hugetlbfs.gigantic_page_threshold_mib << 20;
269 :
270 0 : ulong tile_to_cpu[ FD_TILE_MAX ] = {0};
271 0 : ushort parsed_tile_to_cpu[ FD_TILE_MAX ];
272 : /* Unassigned tiles will be floating, unless auto topology is enabled. */
273 0 : for( ulong i=0UL; i<FD_TILE_MAX; i++ ) parsed_tile_to_cpu[ i ] = USHORT_MAX;
274 :
275 0 : int is_auto_affinity = !strcmp( config->layout.affinity, "auto" );
276 0 : int is_bench_auto_affinity = !strcmp( config->development.bench.affinity, "auto" );
277 :
278 0 : if( FD_UNLIKELY( is_auto_affinity != is_bench_auto_affinity ) ) {
279 0 : FD_LOG_ERR(( "The CPU affinity string in the configuration file under [layout.affinity] and [development.bench.affinity] must all be set to 'auto' or all be set to a specific CPU affinity string." ));
280 0 : }
281 :
282 0 : fd_topo_cpus_t cpus[1];
283 0 : fd_topo_cpus_init( cpus );
284 :
285 0 : ulong affinity_tile_cnt = 0UL;
286 0 : if( FD_LIKELY( !is_auto_affinity ) ) affinity_tile_cnt = fd_topob_parse_affinity_cstr( config->layout.affinity, parsed_tile_to_cpu, 0 );
287 :
288 0 : for( ulong i=0UL; i<affinity_tile_cnt; i++ ) {
289 0 : if( FD_UNLIKELY( parsed_tile_to_cpu[ i ]!=USHORT_MAX && parsed_tile_to_cpu[ i ]>=cpus->cpu_cnt ) )
290 0 : FD_LOG_ERR(( "The CPU affinity string in the configuration file under [layout.affinity] specifies a CPU index of %hu, but the system "
291 0 : "only has %lu CPUs. You should either change the CPU allocations in the affinity string, or increase the number of CPUs "
292 0 : "in the system.",
293 0 : parsed_tile_to_cpu[ i ], cpus->cpu_cnt ));
294 0 : tile_to_cpu[ i ] = fd_ulong_if( parsed_tile_to_cpu[ i ]==USHORT_MAX, ULONG_MAX, (ulong)parsed_tile_to_cpu[ i ] );
295 0 : }
296 :
297 0 : fd_core_subtopo( config, tile_to_cpu );
298 0 : fd_gossip_subtopo( config, tile_to_cpu );
299 :
300 : /* topo, name */
301 0 : fd_topob_wksp( topo, "net_shred" );
302 0 : fd_topob_wksp( topo, "net_repair" );
303 0 : fd_topob_wksp( topo, "net_quic" );
304 :
305 0 : fd_topob_wksp( topo, "shred_out" );
306 0 : fd_topob_wksp( topo, "replay_epoch" );
307 :
308 0 : fd_topob_wksp( topo, "poh_shred" );
309 :
310 0 : fd_topob_wksp( topo, "shred_sign" );
311 0 : fd_topob_wksp( topo, "sign_shred" );
312 :
313 0 : fd_topob_wksp( topo, "repair_sign" );
314 0 : fd_topob_wksp( topo, "sign_repair" );
315 0 : fd_topob_wksp( topo, "rnonce" );
316 0 : fd_topob_wksp( topo, "repair_out" );
317 :
318 0 : fd_topob_wksp( topo, "txsend_out" );
319 :
320 0 : fd_topob_wksp( topo, "shred" );
321 0 : fd_topob_wksp( topo, "repair" );
322 0 : fd_topob_wksp( topo, "fec_sets" );
323 0 : fd_topob_wksp( topo, "snapin_manif" );
324 :
325 0 : fd_topob_wksp( topo, "genesi_out" ); /* mock genesi_out for ipecho */
326 :
327 0 : fd_topob_wksp( topo, "tower_out" ); /* mock tower_out for confirmation msgs. Not needed for any topo except eqvoc. */
328 :
329 0 : #define FOR(cnt) for( ulong i=0UL; i<cnt; i++ )
330 :
331 0 : ulong pending_fec_shreds_depth = fd_ulong_min( fd_ulong_pow2_up( config->tiles.shred.max_pending_shred_sets * FD_REEDSOL_DATA_SHREDS_MAX ), USHORT_MAX + 1 /* dcache max */ );
332 :
333 : /* topo, link_name, wksp_name, depth, mtu, burst */
334 0 : FOR(quic_tile_cnt) fd_topob_link( topo, "quic_net", "net_quic", config->net.ingress_buffer_size, FD_NET_MTU, 1UL );
335 0 : FOR(shred_tile_cnt) fd_topob_link( topo, "shred_net", "net_shred", config->net.ingress_buffer_size, FD_NET_MTU, 1UL );
336 :
337 0 : /**/ fd_topob_link( topo, "replay_epoch", "replay_epoch", 16UL, FD_EPOCH_OUT_MTU, 1UL );
338 :
339 0 : FOR(shred_tile_cnt) fd_topob_link( topo, "shred_sign", "shred_sign", 128UL, 32UL, 1UL );
340 0 : FOR(shred_tile_cnt) fd_topob_link( topo, "sign_shred", "sign_shred", 128UL, 64UL, 1UL );
341 :
342 0 : /**/ fd_topob_link( topo, "repair_net", "net_repair", config->net.ingress_buffer_size, FD_NET_MTU, 1UL );
343 :
344 0 : FOR(shred_tile_cnt) fd_topob_link( topo, "shred_out", "shred_out", pending_fec_shreds_depth, sizeof(fd_shred_message_t), 2UL /* at most 2 msgs per after_frag */ );
345 0 : FOR(sign_tile_cnt-1) fd_topob_link( topo, "repair_sign", "repair_sign", 256UL, FD_REPAIR_MAX_PREIMAGE_SZ, 1UL );
346 0 : FOR(sign_tile_cnt-1) fd_topob_link( topo, "sign_repair", "sign_repair", 128UL, sizeof(fd_ed25519_sig_t), 1UL );
347 :
348 : /**/ fd_topob_link( topo, "repair_out", "repair_out", 128UL, sizeof(fd_fec_complete_t), 1UL );
349 :
350 0 : /**/ fd_topob_link( topo, "poh_shred", "poh_shred", 16384UL, FD_POH_SHRED_MTU, 1UL );
351 :
352 0 : /**/ fd_topob_link( topo, "txsend_out", "txsend_out", 128UL, FD_TXN_MTU, 1UL );
353 :
354 : /**/ fd_topob_link( topo, "snapin_manif", "snapin_manif", 2UL, sizeof(fd_snapshot_manifest_t),1UL );
355 :
356 0 : /**/ fd_topob_link( topo, "genesi_out", "genesi_out", 1UL, FD_GENESIS_TILE_MTU, 1UL );
357 0 : /**/ fd_topob_link( topo, "tower_out", "tower_out", 1024UL, sizeof(fd_tower_msg_t), 1UL );
358 :
359 0 : FOR(net_tile_cnt) fd_topos_net_rx_link( topo, "net_repair", i, config->net.ingress_buffer_size );
360 0 : FOR(net_tile_cnt) fd_topos_net_rx_link( topo, "net_quic", i, config->net.ingress_buffer_size );
361 0 : FOR(net_tile_cnt) fd_topos_net_rx_link( topo, "net_shred", i, config->net.ingress_buffer_size );
362 :
363 : /* topo, tile_name, tile_wksp, metrics_wksp, cpu_idx, is_agave, uses_id_keyswitch, uses_av_keyswitch */
364 0 : FOR(shred_tile_cnt) fd_topob_tile( topo, "shred", "shred", "metric_in", tile_to_cpu[ topo->tile_cnt ], 0, 1, 0 );
365 0 : fd_topo_tile_t * repair_tile = fd_topob_tile( topo, "repair", "repair", "metric_in", tile_to_cpu[ topo->tile_cnt ], 0, 1, 0 );
366 :
367 : /* Setup a shared wksp object for fec sets. */
368 :
369 0 : ulong shred_depth = 65536UL; /* from fdctl/topology.c shred_store link. MAKE SURE TO KEEP IN SYNC. */
370 0 : ulong fec_set_cnt = 2UL*shred_depth + config->tiles.shred.max_pending_shred_sets + 6UL;
371 0 : ulong fec_sets_sz = fec_set_cnt*sizeof(fd_fec_set_t); /* mirrors # of dcache entries in frankendancer */
372 0 : fd_topo_obj_t * fec_sets_obj = setup_topo_fec_sets( topo, "fec_sets", shred_tile_cnt*fec_sets_sz );
373 0 : for( ulong i=0UL; i<shred_tile_cnt; i++ ) {
374 0 : fd_topo_tile_t * shred_tile = &topo->tiles[ fd_topo_find_tile( topo, "shred", i ) ];
375 0 : fd_topob_tile_uses( topo, shred_tile, fec_sets_obj, FD_SHMEM_JOIN_MODE_READ_WRITE );
376 0 : }
377 0 : fd_topob_tile_uses( topo, repair_tile, fec_sets_obj, FD_SHMEM_JOIN_MODE_READ_ONLY );
378 0 : FD_TEST( fd_pod_insertf_ulong( topo->props, fec_sets_obj->id, "fec_sets" ) );
379 :
380 : /* There's another special fseq that's used to communicate the shred
381 : version from the Agave boot path to the shred tile. */
382 0 : fd_topo_obj_t * poh_shred_obj = fd_topob_obj( topo, "fseq", "poh_shred" );
383 0 : fd_topo_tile_t * poh_tile = &topo->tiles[ fd_topo_find_tile( topo, "gossip", 0UL ) ];
384 0 : fd_topob_tile_uses( topo, poh_tile, poh_shred_obj, FD_SHMEM_JOIN_MODE_READ_WRITE );
385 :
386 0 : for( ulong i=0UL; i<shred_tile_cnt; i++ ) {
387 0 : fd_topo_tile_t * shred_tile = &topo->tiles[ fd_topo_find_tile( topo, "shred", i ) ];
388 0 : fd_topob_tile_uses( topo, shred_tile, poh_shred_obj, FD_SHMEM_JOIN_MODE_READ_ONLY );
389 0 : }
390 0 : FD_TEST( fd_pod_insertf_ulong( topo->props, poh_shred_obj->id, "poh_shred" ) );
391 :
392 0 : if( FD_LIKELY( !is_auto_affinity ) ) {
393 0 : if( FD_UNLIKELY( affinity_tile_cnt<topo->tile_cnt ) )
394 0 : FD_LOG_ERR(( "The topology you are using has %lu tiles, but the CPU affinity specified in the config tile as [layout.affinity] only provides for %lu cores. "
395 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 "
396 0 : "the total tile count. You can reduce the tile count by decreasing individual tile counts in the [layout] section of the configuration file.",
397 0 : topo->tile_cnt, affinity_tile_cnt ));
398 0 : if( FD_UNLIKELY( affinity_tile_cnt>topo->tile_cnt ) )
399 0 : FD_LOG_WARNING(( "The topology you are using has %lu tiles, but the CPU affinity specified in the config tile as [layout.affinity] provides for %lu cores. "
400 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 "
401 0 : "individual tile counts in the [layout] section of the configuration file.",
402 0 : topo->tile_cnt, affinity_tile_cnt ));
403 0 : }
404 :
405 : /* topo, tile_name, tile_kind_id, fseq_wksp, link_name, link_kind_id, reliable, polled */
406 0 : for( ulong j=0UL; j<shred_tile_cnt; j++ )
407 0 : fd_topos_tile_in_net( topo, "metric_in", "shred_net", j, FD_TOPOB_UNRELIABLE, FD_TOPOB_POLLED ); /* No reliable consumers of networking fragments, may be dropped or overrun */
408 0 : for( ulong j=0UL; j<quic_tile_cnt; j++ )
409 0 : {fd_topos_tile_in_net( topo, "metric_in", "quic_net", j, FD_TOPOB_UNRELIABLE, FD_TOPOB_POLLED );} /* No reliable consumers of networking fragments, may be dropped or overrun */
410 :
411 0 : /**/ fd_topob_tile_in( topo, "gossip", 0UL, "metric_in", "txsend_out", 0UL, FD_TOPOB_RELIABLE, FD_TOPOB_POLLED );
412 :
413 0 : /**/ fd_topos_tile_in_net( topo, "metric_in", "repair_net", 0UL, FD_TOPOB_UNRELIABLE, FD_TOPOB_POLLED ); /* No reliable consumers of networking fragments, may be dropped or overrun */
414 :
415 0 : FOR(shred_tile_cnt) for( ulong j=0UL; j<net_tile_cnt; j++ )
416 0 : fd_topob_tile_in( topo, "shred", i, "metric_in", "net_shred", j, FD_TOPOB_UNRELIABLE, FD_TOPOB_POLLED ); /* No reliable consumers of networking fragments, may be dropped or overrun */
417 0 : FOR(shred_tile_cnt) fd_topob_tile_in( topo, "shred", i, "metric_in", "poh_shred", 0UL, FD_TOPOB_RELIABLE, FD_TOPOB_POLLED );
418 0 : FOR(shred_tile_cnt) fd_topob_tile_in( topo, "shred", i, "metric_in", "replay_epoch", 0UL, FD_TOPOB_RELIABLE, FD_TOPOB_POLLED );
419 0 : FOR(shred_tile_cnt) fd_topob_tile_in( topo, "shred", i, "metric_in", "gossip_out", 0UL, FD_TOPOB_RELIABLE, FD_TOPOB_POLLED );
420 0 : FOR(shred_tile_cnt) fd_topob_tile_out( topo, "shred", i, "shred_out", i );
421 0 : FOR(shred_tile_cnt) fd_topob_tile_out( topo, "shred", i, "shred_net", i );
422 0 : FOR(shred_tile_cnt) fd_topob_tile_in ( topo, "shred", i, "metric_in", "ipecho_out", 0UL, FD_TOPOB_RELIABLE, FD_TOPOB_POLLED );
423 :
424 : /**/ fd_topob_tile_out( topo, "repair", 0UL, "repair_net", 0UL );
425 :
426 : /* Sign links don't need to be reliable because they are synchronous,
427 : so there's at most one fragment in flight at a time anyway. The
428 : sign links are also not polled by the mux, instead the tiles will
429 : read the sign responses out of band in a dedicated spin loop. */
430 0 : for( ulong i=0UL; i<shred_tile_cnt; i++ ) {
431 0 : /**/ fd_topob_tile_in( topo, "sign", 0UL, "metric_in", "shred_sign", i, FD_TOPOB_UNRELIABLE, FD_TOPOB_POLLED );
432 0 : /**/ fd_topob_tile_out( topo, "shred", i, "shred_sign", i );
433 0 : /**/ fd_topob_tile_in( topo, "shred", i, "metric_in", "sign_shred", i, FD_TOPOB_UNRELIABLE, FD_TOPOB_UNPOLLED );
434 0 : /**/ fd_topob_tile_out( topo, "sign", 0UL, "sign_shred", i );
435 0 : }
436 0 : FOR(gossvf_tile_cnt) fd_topob_tile_in ( topo, "gossvf", i, "metric_in", "replay_epoch", 0UL, FD_TOPOB_RELIABLE, FD_TOPOB_POLLED );
437 :
438 0 : /**/ fd_topob_tile_in ( topo, "gossip", 0UL, "metric_in", "replay_epoch", 0UL, FD_TOPOB_RELIABLE, FD_TOPOB_POLLED );
439 :
440 0 : FOR(net_tile_cnt) fd_topob_tile_in( topo, "repair", 0UL, "metric_in", "net_repair", i, FD_TOPOB_UNRELIABLE, FD_TOPOB_POLLED ); /* No reliable consumers of networking fragments, may be dropped or overrun */
441 0 : /**/ fd_topob_tile_in( topo, "repair", 0UL, "metric_in", "gossip_out", 0UL, FD_TOPOB_RELIABLE, FD_TOPOB_POLLED );
442 0 : fd_topob_tile_in( topo, "repair", 0UL, "metric_in", "snapin_manif", 0UL, FD_TOPOB_RELIABLE, FD_TOPOB_POLLED );
443 0 : FOR(shred_tile_cnt) fd_topob_tile_in( topo, "repair", 0UL, "metric_in", "shred_out", i, FD_TOPOB_RELIABLE, FD_TOPOB_POLLED );
444 0 : FOR(sign_tile_cnt-1) fd_topob_tile_out( topo, "repair", 0UL, "repair_sign", i );
445 0 : FOR(sign_tile_cnt-1) fd_topob_tile_in ( topo, "sign", i+1, "metric_in", "repair_sign", i, FD_TOPOB_RELIABLE, FD_TOPOB_POLLED );
446 0 : FOR(sign_tile_cnt-1) fd_topob_tile_out( topo, "sign", i+1, "sign_repair", i );
447 0 : FOR(sign_tile_cnt-1) fd_topob_tile_in ( topo, "repair", 0UL, "metric_in", "sign_repair", i, FD_TOPOB_UNRELIABLE, FD_TOPOB_POLLED );
448 :
449 : /**/ fd_topob_tile_out( topo, "repair", 0UL, "repair_out", 0UL );
450 0 : /**/ fd_topob_tile_in ( topo, "gossip", 0UL, "metric_in", "sign_gossip", 0UL, FD_TOPOB_UNRELIABLE, FD_TOPOB_UNPOLLED );
451 0 : /**/ fd_topob_tile_in ( topo, "ipecho", 0UL, "metric_in", "genesi_out", 0UL, FD_TOPOB_RELIABLE, FD_TOPOB_POLLED );
452 0 : /**/ fd_topob_tile_in ( topo, "repair", 0UL, "metric_in", "tower_out", 0UL, FD_TOPOB_RELIABLE, FD_TOPOB_POLLED );
453 :
454 : /* Repair and shred share a secret they use to generate the nonces.
455 : It's not super security sensitive, but for good hygiene, we make it
456 : an object. */
457 0 : if( 1 /* just restrict the scope for these variables in this big function */ ) {
458 0 : fd_topo_obj_t * rnonce_ss_obj = fd_topob_obj( topo, "rnonce_ss", "rnonce" );
459 0 : fd_topo_tile_t * repair_tile = &topo->tiles[ fd_topo_find_tile( topo, "repair", 0UL ) ];
460 0 : fd_topob_tile_uses( topo, repair_tile, rnonce_ss_obj, FD_SHMEM_JOIN_MODE_READ_ONLY );
461 0 : for( ulong i=0UL; i<shred_tile_cnt; i++ ) {
462 0 : fd_topo_tile_t * shred_tile = &topo->tiles[ fd_topo_find_tile( topo, "shred", i ) ];
463 0 : fd_topob_tile_uses( topo, shred_tile, rnonce_ss_obj, FD_SHMEM_JOIN_MODE_READ_ONLY );
464 0 : }
465 0 : FD_TEST( fd_pod_insertf_ulong( topo->props, rnonce_ss_obj->id, "rnonce_ss" ) );
466 0 : }
467 :
468 0 : FD_TEST( fd_link_permit_no_producers( topo, "quic_net" ) == quic_tile_cnt );
469 0 : FD_TEST( fd_link_permit_no_producers( topo, "poh_shred" ) == 1UL );
470 0 : FD_TEST( fd_link_permit_no_producers( topo, "txsend_out" ) == 1UL );
471 0 : FD_TEST( fd_link_permit_no_producers( topo, "genesi_out" ) == 1UL );
472 0 : FD_TEST( fd_link_permit_no_producers( topo, "tower_out" ) == 1UL );
473 0 : FD_TEST( fd_link_permit_no_producers( topo, "replay_epoch" ) == 1UL );
474 0 : FD_TEST( fd_link_permit_no_producers( topo, "snapin_manif" ) == 1UL );
475 0 : FD_TEST( fd_link_permit_no_consumers( topo, "net_quic" ) == net_tile_cnt );
476 0 : FD_TEST( fd_link_permit_no_consumers( topo, "repair_out" ) == 1UL );
477 :
478 0 : config->tiles.txsend.txsend_src_port = 0; /* disable txsend */
479 :
480 0 : FOR(net_tile_cnt) fd_topos_net_tile_finish( topo, i );
481 :
482 0 : for( ulong i=0UL; i<topo->tile_cnt; i++ ) {
483 0 : fd_topo_tile_t * tile = &topo->tiles[ i ];
484 0 : fd_topo_configure_tile( tile, config );
485 0 : }
486 :
487 0 : if( FD_UNLIKELY( is_auto_affinity ) ) fd_topob_auto_layout( topo, 0 );
488 :
489 0 : fd_topob_finish( topo, CALLBACKS );
490 :
491 0 : config->topo = *topo;
492 0 : }
493 :
494 : static char *
495 0 : fmt_count( char buf[ static 64 ], ulong count ) {
496 0 : char tmp[ 64 ];
497 0 : if( FD_LIKELY( count<1000UL ) ) FD_TEST( fd_cstr_printf_check( tmp, 64UL, NULL, "%lu", count ) );
498 0 : else if( FD_LIKELY( count<1000000UL ) ) FD_TEST( fd_cstr_printf_check( tmp, 64UL, NULL, "%.1f K", (double)count/1000.0 ) );
499 0 : else if( FD_LIKELY( count<1000000000UL ) ) FD_TEST( fd_cstr_printf_check( tmp, 64UL, NULL, "%.1f M", (double)count/1000000.0 ) );
500 :
501 0 : FD_TEST( fd_cstr_printf_check( buf, 64UL, NULL, "%12s", tmp ) );
502 0 : return buf;
503 0 : }
504 :
505 : static void
506 : print_histogram_buckets( volatile ulong * metrics,
507 : ulong offset,
508 : int converter,
509 : double histmin,
510 : double histmax,
511 0 : char * title ) {
512 0 : fd_histf_t hist[1];
513 :
514 : /* Create histogram structure only to get bucket edges for display */
515 0 : if( FD_LIKELY( converter == FD_METRICS_CONVERTER_SECONDS ) ) {
516 : /* For SLOT_COMPLETE_TIME: min=0.2, max=2.0 seconds */
517 0 : FD_TEST( fd_histf_new( hist, fd_metrics_convert_seconds_to_ticks( histmin ), fd_metrics_convert_seconds_to_ticks( histmax ) ) );
518 0 : } else if( FD_LIKELY( converter == FD_METRICS_CONVERTER_NONE ) ) {
519 : /* For non-time histograms, we'd need the actual min/max values */
520 0 : FD_TEST( fd_histf_new( hist, (ulong)histmin, (ulong)histmax ) );
521 0 : } else {
522 0 : FD_LOG_ERR(( "unknown converter %i", converter ));
523 0 : }
524 :
525 0 : printf( " +---------------------+--------------------+--------------+\n" );
526 0 : printf( " | %-19s | | Count |\n", title );
527 0 : printf( " +---------------------+--------------------+--------------+\n" );
528 :
529 0 : ulong total_count = 0;
530 0 : for( ulong k = 0; k < FD_HISTF_BUCKET_CNT; k++ ) {
531 0 : ulong bucket_count = metrics[ offset + k ];
532 0 : total_count += bucket_count;
533 0 : }
534 :
535 0 : for( ulong k = 0; k < FD_HISTF_BUCKET_CNT; k++ ) {
536 : /* Get individual bucket count directly from metrics array */
537 0 : ulong bucket_count = metrics[ offset + k ];
538 :
539 0 : char * le_str;
540 0 : char le_buf[ 64 ];
541 0 : if( FD_UNLIKELY( k == FD_HISTF_BUCKET_CNT - 1UL ) ) {
542 0 : le_str = "+Inf";
543 0 : } else {
544 0 : ulong edge = fd_histf_right( hist, k );
545 0 : if( FD_LIKELY( converter == FD_METRICS_CONVERTER_SECONDS ) ) {
546 0 : double edgef = fd_metrics_convert_ticks_to_seconds( edge - 1 );
547 0 : FD_TEST( fd_cstr_printf_check( le_buf, sizeof( le_buf ), NULL, "%.3f", edgef ) );
548 0 : } else {
549 0 : FD_TEST( fd_cstr_printf_check( le_buf, sizeof( le_buf ), NULL, "%.3f", (double)(edge - 1) / 1000000.0 ) );
550 0 : }
551 0 : le_str = le_buf;
552 0 : }
553 :
554 0 : char count_buf[ 64 ];
555 0 : fmt_count( count_buf, bucket_count );
556 :
557 : /* Match visual bar length to the %-18s display column width. */
558 0 : char bar_buf[ 19 ];
559 0 : ulong bar_max = sizeof( bar_buf ) - 1UL;
560 0 : if( bucket_count > 0 && total_count > 0 ) {
561 0 : ulong bar_length = (bucket_count * bar_max) / total_count;
562 0 : if( bar_length == 0 ) bar_length = 1;
563 0 : if( bar_length > bar_max ) bar_length = bar_max;
564 0 : for( ulong i = 0; i < bar_length; i++ ) { bar_buf[ i ] = '|'; }
565 0 : bar_buf[ bar_length ] = '\0';
566 0 : } else {
567 0 : bar_buf[ 0 ] = '\0';
568 0 : }
569 :
570 0 : printf( " | %-19s | %-18s | %s |\n", le_str, bar_buf, count_buf );
571 0 : }
572 :
573 : /* Print sum and total count */
574 0 : char sum_buf[ 64 ];
575 0 : char avg_buf[ 64 ];
576 0 : if( FD_LIKELY( converter == FD_METRICS_CONVERTER_SECONDS ) ) {
577 0 : double sumf = fd_metrics_convert_ticks_to_seconds( metrics[ offset + FD_HISTF_BUCKET_CNT ] );
578 0 : FD_TEST( fd_cstr_printf_check( sum_buf, sizeof( sum_buf ), NULL, "%.6f", sumf ) );
579 0 : double avg = sumf / (double)total_count;
580 0 : FD_TEST( fd_cstr_printf_check( avg_buf, sizeof( avg_buf ), NULL, "%.6f", avg ) );
581 0 : } else {
582 0 : FD_TEST( fd_cstr_printf_check( sum_buf, sizeof( sum_buf ), NULL, "%lu", metrics[ offset + FD_HISTF_BUCKET_CNT ] ));
583 0 : }
584 :
585 0 : printf( " +---------------------+--------------------+---------------+\n" );
586 0 : printf( " | Sum: %-14s | Count: %-11lu | Avg: %-8s |\n", sum_buf, total_count, avg_buf );
587 0 : printf( " +---------------------+--------------------+---------------+\n" );
588 0 : }
589 :
590 : static fd_slot_metrics_t temp_slots[ FD_CATCHUP_METRICS_MAX ];
591 :
592 : static void
593 0 : print_catchup_slots( fd_wksp_t * repair_tile_wksp, ctx_t * repair_ctx, int verbose, int sort_by_slot ) {
594 0 : fd_repair_metrics_t * catchup = repair_ctx->slot_metrics;
595 0 : ulong catchup_gaddr = fd_wksp_gaddr_fast( repair_ctx->wksp, catchup );
596 0 : fd_repair_metrics_t * catchup_table = (fd_repair_metrics_t *)fd_wksp_laddr( repair_tile_wksp, catchup_gaddr );
597 0 : if( FD_LIKELY( sort_by_slot ) ) {
598 0 : fd_repair_metrics_print_sorted( catchup_table, verbose, temp_slots );
599 0 : } else {
600 0 : fd_repair_metrics_print( catchup_table, verbose );
601 0 : }
602 0 : }
603 :
604 : static fd_location_info_t * location_table;
605 : static fd_pubkey_t peers_copy[ FD_REPAIR_PEER_MAX];
606 :
607 : static ulong
608 0 : sort_peers_by_latency( fd_policy_peer_map_t * active_table, fd_policy_peer_dlist_t * peers_dlist, fd_policy_peer_dlist_t * peers_wlist, fd_policy_peer_t * peers_arr ) {
609 0 : ulong i = 0;
610 0 : fd_policy_peer_dlist_iter_t iter = fd_policy_peer_dlist_iter_fwd_init( peers_dlist, peers_arr );
611 0 : while( !fd_policy_peer_dlist_iter_done( iter, peers_dlist, peers_arr ) ) {
612 0 : fd_policy_peer_t * peer = fd_policy_peer_dlist_iter_ele( iter, peers_dlist, peers_arr );
613 0 : if( FD_UNLIKELY( !peer ) ) break;
614 0 : peers_copy[ i++ ] = peer->key;
615 0 : if( FD_UNLIKELY( i >= FD_REPAIR_PEER_MAX ) ) break;
616 0 : iter = fd_policy_peer_dlist_iter_fwd_next( iter, peers_dlist, peers_arr );
617 0 : }
618 0 : ulong fast_cnt = i;
619 0 : iter = fd_policy_peer_dlist_iter_fwd_init( peers_wlist, peers_arr );
620 0 : while( !fd_policy_peer_dlist_iter_done( iter, peers_wlist, peers_arr ) ) {
621 0 : fd_policy_peer_t * peer = fd_policy_peer_dlist_iter_ele( iter, peers_wlist, peers_arr );
622 0 : if( FD_UNLIKELY( !peer ) ) break;
623 0 : peers_copy[ i++ ] = peer->key;
624 0 : if( FD_UNLIKELY( i >= FD_REPAIR_PEER_MAX ) ) break;
625 0 : iter = fd_policy_peer_dlist_iter_fwd_next( iter, peers_wlist, peers_arr );
626 0 : }
627 0 : FD_LOG_NOTICE(( "Fast peers cnt: %lu. Slow peers cnt: %lu.", fast_cnt, i - fast_cnt ));
628 :
629 0 : ulong peer_cnt = i;
630 0 : for( uint i = 0; i < peer_cnt - 1; i++ ) {
631 0 : int swapped = 0;
632 0 : for( uint j = 0; j < peer_cnt - 1 - i; j++ ) {
633 0 : fd_policy_peer_t const * active_j = fd_policy_peer_map_ele_query( active_table, &peers_copy[ j ], NULL, peers_arr );
634 0 : fd_policy_peer_t const * active_j1 = fd_policy_peer_map_ele_query( active_table, &peers_copy[ j + 1 ], NULL, peers_arr );
635 :
636 : /* Skip peers with no responses */
637 0 : double latency_j = 10e9;
638 0 : double latency_j1 = 10e9;
639 0 : if( FD_LIKELY( active_j && active_j->res_cnt > 0 ) ) latency_j = ((double)active_j->total_lat / (double)active_j->res_cnt);
640 0 : if( FD_LIKELY( active_j1 && active_j1->res_cnt > 0 ) ) latency_j1 = ((double)active_j1->total_lat / (double)active_j1->res_cnt);
641 :
642 : /* Swap if j has higher latency than j+1 */
643 0 : if( latency_j > latency_j1 ) {
644 0 : fd_pubkey_t temp = peers_copy[ j ];
645 0 : peers_copy[ j ] = peers_copy[ j + 1 ];
646 0 : peers_copy[ j + 1 ] = temp;
647 0 : swapped = 1;
648 0 : }
649 0 : }
650 0 : if( !swapped ) break;
651 0 : }
652 0 : return peer_cnt;
653 0 : }
654 :
655 : static void
656 0 : print_peer_location_latency( fd_wksp_t * repair_tile_wksp, ctx_t * tile_ctx ) {
657 0 : ulong policy_gaddr = fd_wksp_gaddr_fast( tile_ctx->wksp, tile_ctx->policy );
658 0 : fd_policy_t * policy = fd_wksp_laddr ( repair_tile_wksp, policy_gaddr );
659 0 : ulong peermap_gaddr = fd_wksp_gaddr_fast( tile_ctx->wksp, policy->peers.map );
660 0 : ulong peerarr_gaddr = fd_wksp_gaddr_fast( tile_ctx->wksp, policy->peers.pool );
661 0 : ulong peerlst_gaddr = fd_wksp_gaddr_fast( tile_ctx->wksp, policy->peers.fast );
662 0 : ulong peerwst_gaddr = fd_wksp_gaddr_fast( tile_ctx->wksp, policy->peers.slow );
663 0 : fd_policy_peer_map_t * peers_map = (fd_policy_peer_map_t *) fd_wksp_laddr( repair_tile_wksp, peermap_gaddr );
664 0 : fd_policy_peer_dlist_t * peers_dlist = (fd_policy_peer_dlist_t *)fd_wksp_laddr( repair_tile_wksp, peerlst_gaddr );
665 0 : fd_policy_peer_dlist_t * peers_wlist = (fd_policy_peer_dlist_t *)fd_wksp_laddr( repair_tile_wksp, peerwst_gaddr );
666 0 : fd_policy_peer_t * peers_arr = (fd_policy_peer_t *) fd_wksp_laddr( repair_tile_wksp, peerarr_gaddr );
667 :
668 0 : ulong peer_cnt = sort_peers_by_latency( peers_map, peers_dlist, peers_wlist, peers_arr );
669 0 : printf("\nPeer Location/Latency Information\n");
670 0 : printf( " | %-46s | %-7s | %-8s | %-8s | %-7s | %-7s | %-12s | %s\n", "Pubkey", "Req Cnt", "Req B/s", "Rx B/s", "Rx Rate", "Avg Latency", "Ewma Latency", "Location Info" );
671 0 : for( uint i = 0; i < peer_cnt; i++ ) {
672 0 : fd_policy_peer_t const * active = fd_policy_peer_map_ele_query( peers_map, &peers_copy[ i ], NULL, peers_arr );
673 0 : if( FD_LIKELY( active && active->res_cnt > 0 ) ) {
674 0 : fd_location_info_t * info = fd_location_table_query( location_table, active->ip4, NULL );
675 0 : char * geolocation = info ? info->location : "";
676 0 : double peer_bps = (double)(active->res_cnt * FD_SHRED_MIN_SZ) / ((double)(active->last_resp_ts - active->first_resp_ts) / 1e9);
677 0 : double req_bps = (double)active->req_cnt * 202 / ((double)(active->last_req_ts - active->first_req_ts) / 1e9);
678 0 : FD_BASE58_ENCODE_32_BYTES( active->key.key, key_b58 );
679 0 : printf( "%-5u | %-46s | %-7lu | %-8.2f | %-8.2f | %-7.2f | %10.3fms | %10.3fms | %s\n", i, key_b58, active->req_cnt, req_bps, peer_bps, (double)active->res_cnt / (double)active->req_cnt, ((double)active->total_lat / (double)active->res_cnt) / 1e6, (double)active->ewma_lat / 1e6, geolocation );
680 0 : }
681 0 : }
682 0 : printf("\n");
683 0 : fflush( stdout );
684 0 : }
685 :
686 : static void
687 0 : read_iptable( char * iptable_path, fd_location_info_t * location_table ) {
688 0 : int iptable_fd = open( iptable_path, O_RDONLY );
689 0 : if( FD_UNLIKELY( iptable_fd<0 ) ) return;
690 :
691 : /* read iptable line by line */
692 0 : if( FD_LIKELY( iptable_fd>=0 ) ) {
693 0 : char line[ 256 ];
694 0 : uchar istream_buf[256];
695 0 : fd_io_buffered_istream_t istream[1];
696 0 : fd_io_buffered_istream_init( istream, iptable_fd, istream_buf, sizeof(istream_buf) );
697 0 : for(;;) {
698 0 : int err;
699 0 : if( !fd_io_fgets( line, sizeof(line), istream, &err ) ) break;
700 0 : fd_location_info_t location_info;
701 0 : sscanf( line, "%lu %[^\n]", &location_info.ip4_addr, location_info.location );
702 0 : fd_location_info_t * info = fd_location_table_insert( location_table, location_info.ip4_addr );
703 0 : if( FD_UNLIKELY( info==NULL ) ) break;
704 0 : memcpy( info->location, location_info.location, sizeof(info->location) );
705 0 : }
706 0 : }
707 0 : }
708 :
709 : static void
710 : print_tile_metrics( volatile ulong * shred_metrics,
711 : volatile ulong * repair_metrics,
712 : volatile ulong * repair_metrics_prev, /* for diffing metrics */
713 : volatile ulong ** repair_net_links,
714 : volatile ulong ** net_shred_links,
715 : ulong net_tile_cnt,
716 : ulong * last_sent_cnt,
717 : long last_print_ts,
718 0 : long now ) {
719 0 : char buf2[ 64 ];
720 0 : ulong rcvd = shred_metrics [ MIDX( COUNTER, SHRED, SHRED_REPAIR_RX ) ];
721 0 : ulong sent = repair_metrics[ MIDX( COUNTER, REPAIR, REQUEST_TX_NEEDED_WINDOW ) ] +
722 0 : repair_metrics[ MIDX( COUNTER, REPAIR, REQUEST_TX_NEEDED_HIGHEST_WINDOW ) ] +
723 0 : repair_metrics[ MIDX( COUNTER, REPAIR, REQUEST_TX_NEEDED_ORPHAN ) ];
724 0 : printf(" Requests received: (%lu/%lu) %.1f%% \n", rcvd, sent, (double)rcvd / (double)sent * 100.0 );
725 0 : printf( " +---------------+--------------+\n" );
726 0 : printf( " | Request Type | Count |\n" );
727 0 : printf( " +---------------+--------------+\n" );
728 0 : printf( " | Orphan | %s |\n", fmt_count( buf2, repair_metrics[ MIDX( COUNTER, REPAIR, REQUEST_TX_NEEDED_ORPHAN ) ] ) );
729 0 : printf( " | HighestWindow | %s |\n", fmt_count( buf2, repair_metrics[ MIDX( COUNTER, REPAIR, REQUEST_TX_NEEDED_HIGHEST_WINDOW ) ] ) );
730 0 : printf( " | Index | %s |\n", fmt_count( buf2, repair_metrics[ MIDX( COUNTER, REPAIR, REQUEST_TX_NEEDED_WINDOW ) ] ) );
731 0 : printf( " +---------------+--------------+\n" );
732 0 : printf( " Send Pkt Rate: %s pps\n", fmt_count( buf2, (ulong)((sent - *last_sent_cnt)*1e9L / (now - last_print_ts) ) ) );
733 0 : *last_sent_cnt = sent;
734 :
735 : /* Sum overrun across all net tiles connected to repair_net */
736 0 : ulong total_overrun = repair_net_links[0][ MIDX( COUNTER, LINK, FRAG_POLLING_OVERRUN ) ]; /* coarse double counting prevention */
737 0 : ulong total_consumed = 0UL;
738 0 : for( ulong i = 0UL; i < net_tile_cnt; i++ ) {
739 0 : volatile ulong * ovar_net_metrics = repair_net_links[i];
740 0 : total_overrun += ovar_net_metrics[ MIDX( COUNTER, LINK, FRAG_READING_OVERRUN ) ];
741 0 : total_consumed += ovar_net_metrics[ MIDX( COUNTER, LINK, FRAG_CONSUMED ) ]; /* consumed is incremented after after_frag is called */
742 0 : }
743 0 : printf( " Outgoing requests overrun: %s\n", fmt_count( buf2, total_overrun ) );
744 0 : printf( " Outgoing requests consumed: %s\n", fmt_count( buf2, total_consumed ) );
745 :
746 0 : total_overrun = net_shred_links[0][ MIDX( COUNTER, LINK, FRAG_READING_OVERRUN ) ];
747 0 : total_consumed = 0UL;
748 0 : for( ulong i = 0UL; i < net_tile_cnt; i++ ) {
749 0 : volatile ulong * ovar_net_metrics = net_shred_links[i];
750 0 : total_overrun += ovar_net_metrics[ MIDX( COUNTER, LINK, FRAG_READING_OVERRUN ) ];
751 0 : total_consumed += ovar_net_metrics[ MIDX( COUNTER, LINK, FRAG_CONSUMED ) ]; /* shred frag filtering happens manually in after_frag, so no need to index every shred_tile. */
752 0 : }
753 :
754 0 : printf( " Incoming shreds overrun: %s\n", fmt_count( buf2, total_overrun ) );
755 0 : printf( " Incoming shreds consumed: %s\n", fmt_count( buf2, total_consumed ) );
756 :
757 0 : print_histogram_buckets( repair_metrics,
758 0 : MIDX( HISTOGRAM, REPAIR, RESPONSE_LATENCY_NANOS ),
759 0 : FD_METRICS_CONVERTER_NONE,
760 0 : FD_METRICS_HISTOGRAM_REPAIR_RESPONSE_LATENCY_NANOS_MIN,
761 0 : FD_METRICS_HISTOGRAM_REPAIR_RESPONSE_LATENCY_NANOS_MAX,
762 0 : "Response Latency" );
763 :
764 0 : printf(" Repair Peers: %lu\n", repair_metrics[ MIDX( COUNTER, REPAIR, PEER_REQUESTED ) ] );
765 0 : printf(" Shreds rejected (no stakes): %lu\n", shred_metrics[ MIDX( COUNTER, SHRED, SHRED_PROCESSED ) ] );
766 : /* Print histogram buckets similar to Prometheus format */
767 0 : print_histogram_buckets( repair_metrics,
768 0 : MIDX( HISTOGRAM, REPAIR, SLOT_COMPLETE_DURATION_SECONDS ),
769 0 : FD_METRICS_CONVERTER_SECONDS,
770 0 : FD_METRICS_HISTOGRAM_REPAIR_SLOT_COMPLETE_DURATION_SECONDS_MIN,
771 0 : FD_METRICS_HISTOGRAM_REPAIR_SLOT_COMPLETE_DURATION_SECONDS_MAX,
772 0 : "Slot Complete Time" );
773 :
774 0 : #define DIFFX(METRIC) repair_metrics[ MIDX( COUNTER, TILE, METRIC ) ] - repair_metrics_prev[ MIDX( COUNTER, TILE, METRIC ) ]
775 0 : ulong hkeep_ticks = DIFFX(REGIME_DURATION_NANOS_CAUGHT_UP_HOUSEKEEPING) + DIFFX(REGIME_DURATION_NANOS_PROCESSING_HOUSEKEEPING) + DIFFX(REGIME_DURATION_NANOS_BACKPRESSURE_HOUSEKEEPING);
776 0 : ulong busy_ticks = DIFFX(REGIME_DURATION_NANOS_PROCESSING_PREFRAG) + DIFFX(REGIME_DURATION_NANOS_PROCESSING_POSTFRAG ) + DIFFX(REGIME_DURATION_NANOS_CAUGHT_UP_PREFRAG);
777 0 : ulong caught_up_ticks = DIFFX(REGIME_DURATION_NANOS_CAUGHT_UP_POSTFRAG);
778 0 : ulong backpressure_ticks = DIFFX(REGIME_DURATION_NANOS_BACKPRESSURE_PREFRAG);
779 0 : ulong total_ticks = hkeep_ticks + busy_ticks + caught_up_ticks + backpressure_ticks;
780 :
781 0 : printf( " Repair Hkeep: %.1f %% Busy: %.1f %% Idle: %.1f %% Backp: %0.1f %%\n",
782 0 : (double)hkeep_ticks/(double)total_ticks*100.0,
783 0 : (double)busy_ticks/(double)total_ticks*100.0,
784 0 : (double)caught_up_ticks/(double)total_ticks*100.0,
785 0 : (double)backpressure_ticks/(double)total_ticks*100.0 );
786 0 : #undef DIFFX
787 0 : fflush( stdout );
788 :
789 0 : printf( " Block failed insert: %lu\n", repair_metrics[ MIDX( COUNTER, REPAIR, BLOCK_INSERT_FAILED ) ] );
790 0 : printf( " Block evicted: %lu\n", repair_metrics[ MIDX( COUNTER, REPAIR, BLOCK_EVICTED ) ] );
791 0 : printf( " slot evicted: %lu\n", repair_metrics[ MIDX( GAUGE, REPAIR, SLOT_LAST_EVICTED ) ] );
792 0 : printf( " slot evicted by: %lu\n", repair_metrics[ MIDX( GAUGE, REPAIR, SLOT_LAST_EVICTION_CAUSE ) ] );
793 0 : printf( " slot failed insert: %lu\n", repair_metrics[ MIDX( GAUGE, REPAIR, SLOT_LAST_INSERT_FAILED ) ] );
794 0 : for( ulong i=0UL; i<FD_METRICS_TOTAL_SZ/sizeof(ulong); i++ ) repair_metrics_prev[ i ] = repair_metrics[ i ];
795 0 : }
796 :
797 : static void
798 : repair_ctx_wksp( args_t * args,
799 : config_t * config,
800 : ctx_t ** repair_ctx,
801 0 : fd_topo_wksp_t ** repair_wksp ) {
802 0 : (void)args;
803 :
804 0 : fd_topo_t * topo = &config->topo;
805 0 : ulong wksp_id = fd_topo_find_wksp( topo, "repair" );
806 0 : if( FD_UNLIKELY( wksp_id==ULONG_MAX ) ) FD_LOG_ERR(( "repair workspace not found" ));
807 :
808 0 : fd_topo_wksp_t * _repair_wksp = &topo->workspaces[ wksp_id ];
809 :
810 0 : ulong tile_id = fd_topo_find_tile( topo, "repair", 0UL );
811 0 : if( FD_UNLIKELY( tile_id==ULONG_MAX ) ) FD_LOG_ERR(( "repair tile not found" ));
812 :
813 0 : fd_topo_join_workspace( topo, _repair_wksp, FD_SHMEM_JOIN_MODE_READ_ONLY, FD_TOPO_CORE_DUMP_LEVEL_DISABLED );
814 :
815 : /* Access the repair tile scratch memory where repair_tile_ctx is stored */
816 0 : fd_topo_tile_t * tile = &topo->tiles[ tile_id ];
817 0 : void * scratch = fd_topo_obj_laddr( &config->topo, tile->tile_obj_id );
818 0 : if( FD_UNLIKELY( !scratch ) ) FD_LOG_ERR(( "Failed to access repair tile scratch memory" ));
819 :
820 0 : FD_SCRATCH_ALLOC_INIT( l, scratch );
821 0 : ctx_t * _repair_ctx = FD_SCRATCH_ALLOC_APPEND( l, alignof(ctx_t), sizeof(ctx_t) );
822 :
823 0 : *repair_ctx = _repair_ctx;
824 0 : *repair_wksp = _repair_wksp;
825 0 : }
826 :
827 : static void
828 : repair_cmd_fn_catchup( args_t * args,
829 0 : config_t * config ) {
830 :
831 0 : memset( &config->topo, 0, sizeof(config->topo) );
832 0 : repair_topo( config );
833 :
834 0 : fd_topo_print_log( 1, &config->topo );
835 :
836 0 : args_t configure_args = {
837 0 : .configure.command = CONFIGURE_CMD_INIT,
838 0 : };
839 0 : for( ulong i=0UL; STAGES[ i ]; i++ ) {
840 0 : configure_args.configure.stages[ i ] = STAGES[ i ];
841 0 : }
842 0 : configure_cmd_fn( &configure_args, config );
843 0 : if( 0==strcmp( config->net.provider, "xdp" ) ) {
844 0 : fd_topo_install_xdp_simple( &config->topo, config->net.bind_address_parsed );
845 0 : }
846 0 : run_firedancer_init( config, 1, 0 );
847 :
848 0 : fd_topo_join_workspaces( &config->topo, FD_SHMEM_JOIN_MODE_READ_WRITE, FD_TOPO_CORE_DUMP_LEVEL_DISABLED );
849 :
850 0 : fd_topo_fill( &config->topo );
851 :
852 0 : repair_load_manifest( &config->topo, args->repair.manifest_path );
853 :
854 : /* Access repair workspace memory and metrics */
855 :
856 0 : ulong repair_tile_idx = fd_topo_find_tile( &config->topo, "repair", 0UL );
857 0 : ulong shred_tile_idx = fd_topo_find_tile( &config->topo, "shred", 0UL );
858 0 : FD_TEST( repair_tile_idx!=ULONG_MAX );
859 0 : FD_TEST( shred_tile_idx !=ULONG_MAX );
860 0 : fd_topo_tile_t * repair_tile = &config->topo.tiles[ repair_tile_idx ];
861 0 : fd_topo_tile_t * shred_tile = &config->topo.tiles[ shred_tile_idx ];
862 :
863 0 : fd_topo_wksp_t * repair_wksp;
864 0 : ctx_t * repair_ctx;
865 0 : repair_ctx_wksp( args, config, &repair_ctx, &repair_wksp );
866 :
867 0 : volatile ulong * shred_metrics = fd_metrics_tile( shred_tile->metrics );
868 0 : volatile ulong * repair_metrics = fd_metrics_tile( repair_tile->metrics );
869 0 : FD_TEST( repair_metrics );
870 0 : ulong * repair_metrics_prev = aligned_alloc( 8UL, sizeof(ulong) * FD_METRICS_TOTAL_SZ );
871 0 : FD_TEST( repair_metrics_prev );
872 0 : memset( repair_metrics_prev, 0, sizeof(ulong) * FD_METRICS_TOTAL_SZ );
873 :
874 : /* Collect link metrics */
875 :
876 : /* Collect all net tiles and their repair_net link metrics */
877 0 : ulong net_cnt = config->layout.net_tile_count;
878 0 : volatile ulong ** repair_net_links = aligned_alloc( 8UL, net_cnt * sizeof(volatile ulong*) );
879 0 : volatile ulong ** net_shred_links = aligned_alloc( 8UL, net_cnt * sizeof(volatile ulong*) );
880 0 : FD_TEST( repair_net_links );
881 0 : FD_TEST( net_shred_links );
882 :
883 0 : for( ulong i = 0UL; i < net_cnt; i++ ) {
884 0 : ulong tile_idx = fd_topo_find_tile( &config->topo, "net", i );
885 0 : if( FD_UNLIKELY( tile_idx == ULONG_MAX ) ) FD_LOG_ERR(( "net tile %lu not found", i ));
886 0 : fd_topo_tile_t * tile = &config->topo.tiles[ tile_idx ];
887 :
888 0 : ulong repair_net_in_idx = fd_topo_find_tile_in_link( &config->topo, tile, "repair_net", 0UL );
889 0 : if( FD_UNLIKELY( repair_net_in_idx == ULONG_MAX ) ) FD_LOG_ERR(( "repair_net link not found for net tile %lu", i ));
890 0 : FD_TEST( tile->metrics );
891 0 : repair_net_links[i] = fd_metrics_link_in( tile->metrics, repair_net_in_idx );
892 0 : FD_TEST( repair_net_links[i] );
893 :
894 : /* process all net_shred links */
895 0 : ulong shred_tile_idx = fd_topo_find_tile( &config->topo, "shred", 0 );
896 0 : if( FD_UNLIKELY( shred_tile_idx == ULONG_MAX ) ) FD_LOG_ERR(( "shred tile 0 not found" ));
897 0 : fd_topo_tile_t * shred_tile = &config->topo.tiles[ shred_tile_idx ];
898 :
899 0 : ulong shred_out_in_idx = fd_topo_find_tile_in_link( &config->topo, shred_tile, "net_shred", i );
900 0 : if( FD_UNLIKELY( shred_out_in_idx == ULONG_MAX ) ) FD_LOG_ERR(( "net_shred link not found for shred tile 0" ));
901 0 : FD_TEST( shred_tile->metrics );
902 0 : net_shred_links[i] = fd_metrics_link_in( shred_tile->metrics, shred_out_in_idx );
903 0 : FD_TEST( net_shred_links[i] );
904 0 : }
905 :
906 0 : FD_LOG_NOTICE(( "Repair catchup run" ));
907 :
908 0 : ulong shred_out_link_idx = fd_topo_find_link( &config->topo, "shred_out", 0UL );
909 0 : FD_TEST( shred_out_link_idx!=ULONG_MAX );
910 0 : fd_topo_link_t * shred_out_link = &config->topo.links[ shred_out_link_idx ];
911 0 : fd_frag_meta_t * shred_out_mcache = shred_out_link->mcache;
912 0 : void * shred_out_dcache = config->topo.workspaces[ config->topo.objs[ shred_out_link->dcache_obj_id ].wksp_id ].wksp;
913 :
914 0 : ulong turbine_slot0 = 0;
915 0 : long last_print = fd_log_wallclock();
916 0 : ulong last_sent = 0UL;
917 :
918 0 : fd_topo_run_single_process( &config->topo, 0, config->uid, config->gid, fdctl_tile_run );
919 0 : for(;;) {
920 :
921 0 : if( FD_UNLIKELY( !turbine_slot0 ) ) {
922 0 : fd_frag_meta_t * frag = &shred_out_mcache[0]; /* hack to get first frag */
923 0 : if ( frag->sz > 0 ) {
924 0 : uchar * shred_out_chunk = fd_chunk_to_laddr( shred_out_dcache, frag->chunk );
925 0 : fd_shred_base_t * shred_out_shred = (fd_shred_base_t *)fd_type_pun( shred_out_chunk );
926 0 : turbine_slot0 = shred_out_shred->shred.slot;
927 0 : FD_LOG_NOTICE(("turbine_slot0: %lu", turbine_slot0));
928 0 : }
929 0 : }
930 :
931 : /* print metrics */
932 :
933 0 : long now = fd_log_wallclock();
934 0 : int catchup_finished = 0;
935 0 : if( FD_UNLIKELY( now - last_print > 1e9L ) ) {
936 0 : print_tile_metrics( shred_metrics, repair_metrics, repair_metrics_prev, repair_net_links, net_shred_links, net_cnt, &last_sent, last_print, now );
937 0 : ulong slots_behind = turbine_slot0 > repair_metrics[ MIDX( GAUGE, REPAIR, SLOT_HIGHEST_REPAIRED ) ] ? turbine_slot0 - repair_metrics[ MIDX( GAUGE, REPAIR, SLOT_HIGHEST_REPAIRED ) ] : 0;
938 0 : printf(" Repaired slots: %lu/%lu (slots behind: %lu)\n", repair_metrics[ MIDX( GAUGE, REPAIR, SLOT_HIGHEST_REPAIRED ) ], turbine_slot0, slots_behind );
939 0 : if( turbine_slot0 && !slots_behind ) {
940 0 : catchup_finished = 1;
941 0 : }
942 0 : printf("\n");
943 0 : fflush( stdout );
944 0 : last_print = now;
945 0 : }
946 :
947 0 : if( FD_UNLIKELY( catchup_finished ) ) {
948 : /* repair cmd owned memory */
949 0 : location_table = fd_location_table_join( fd_location_table_new( location_table_mem ) );
950 0 : read_iptable( args->repair.iptable_path, location_table );
951 0 : print_peer_location_latency( repair_wksp->wksp, repair_ctx );
952 0 : print_catchup_slots( repair_wksp->wksp, repair_ctx, 0, 1 );
953 0 : FD_LOG_NOTICE(("Catchup to slot %lu completed successfully", turbine_slot0));
954 0 : fd_sys_util_exit_group( 0 );
955 0 : }
956 0 : }
957 0 : }
958 :
959 : /* Tests equivocation detection & repair path. */
960 : static void
961 : repair_cmd_fn_eqvoc( args_t * args,
962 0 : config_t * config ) {
963 0 : (void)args;
964 0 : memset( &config->topo, 0, sizeof(config->topo) );
965 0 : repair_topo( config );
966 :
967 0 : FD_LOG_NOTICE(( "Repair eqvoc testing init" ));
968 0 : fd_topo_print_log( 1, &config->topo );
969 :
970 0 : args_t configure_args = { .configure.command = CONFIGURE_CMD_INIT, };
971 0 : for( ulong i=0UL; STAGES[ i ]; i++ ) configure_args.configure.stages[ i ] = STAGES[ i ];
972 0 : configure_cmd_fn( &configure_args, config );
973 0 : if( 0==strcmp( config->net.provider, "xdp" ) ) fd_topo_install_xdp_simple( &config->topo, config->net.bind_address_parsed );
974 :
975 0 : run_firedancer_init( config, 1, 0 );
976 0 : fd_topo_join_workspaces( &config->topo, FD_SHMEM_JOIN_MODE_READ_WRITE, FD_TOPO_CORE_DUMP_LEVEL_DISABLED );
977 0 : fd_topo_fill( &config->topo );
978 :
979 0 : repair_load_manifest( &config->topo, args->repair.manifest_path );
980 :
981 0 : ulong repair_tile_idx = fd_topo_find_tile( &config->topo, "repair", 0UL );
982 0 : fd_topo_tile_t * repair_tile = &config->topo.tiles[ repair_tile_idx ];
983 0 : volatile ulong * repair_metrics = fd_metrics_tile( repair_tile->metrics );
984 :
985 0 : void * scratch = fd_topo_obj_laddr( &config->topo, repair_tile->tile_obj_id );
986 0 : if( FD_UNLIKELY( !scratch ) ) FD_LOG_ERR(( "Failed to access repair tile scratch memory" ));
987 0 : FD_SCRATCH_ALLOC_INIT( l, scratch );
988 0 : ctx_t * repair_ctx = FD_SCRATCH_ALLOC_APPEND( l, alignof(ctx_t), sizeof(ctx_t) );
989 0 : (void)repair_ctx;
990 :
991 : /* read tower_out mcache dcache */
992 0 : ulong tower_out_link_idx = fd_topo_find_link( &config->topo, "tower_out", 0UL );
993 0 : FD_TEST( tower_out_link_idx!=ULONG_MAX );
994 0 : fd_topo_link_t * tower_out_link = &config->topo.links[ tower_out_link_idx ];
995 0 : fd_frag_meta_t * tower_out_mcache = tower_out_link->mcache;
996 0 : fd_wksp_t * tower_out_mem = config->topo.workspaces[ config->topo.objs[ tower_out_link->dcache_obj_id ].wksp_id ].wksp;
997 0 : ulong tower_out_chunk0 = fd_dcache_compact_chunk0( tower_out_mem, tower_out_link->dcache );
998 0 : ulong tower_out_wmark = fd_dcache_compact_wmark( tower_out_mem, tower_out_link->dcache, tower_out_link->mtu );
999 0 : ulong tower_out_chunk = tower_out_chunk0;
1000 :
1001 0 : fd_topo_run_single_process( &config->topo, 0, config->uid, config->gid, fdctl_tile_run );
1002 0 : int confirmed = 0;
1003 0 : for(;;) {
1004 : /* publish a confirmation on tower_out */
1005 0 : if( FD_UNLIKELY( !confirmed && repair_metrics[ MIDX( GAUGE, REPAIR, SLOT_HIGHEST_REPAIRED ) ] != 0 ) ) {
1006 0 : fd_tower_slot_confirmed_t * msg = fd_chunk_to_laddr( tower_out_mem, tower_out_chunk );
1007 0 : FD_LOG_NOTICE(( "publishing confirmation for slot %lu", msg->slot ));
1008 0 : fd_mcache_publish( tower_out_mcache, tower_out_link->depth, 0, FD_TOWER_SIG_SLOT_CONFIRMED, tower_out_chunk, sizeof(fd_tower_slot_confirmed_t), 0, 0, 0 );
1009 0 : tower_out_chunk = fd_dcache_compact_next( tower_out_chunk, sizeof(fd_tower_slot_confirmed_t), tower_out_chunk0, tower_out_wmark );
1010 0 : confirmed = 1;
1011 0 : }
1012 0 : sleep( 1 );
1013 0 : }
1014 0 : }
1015 :
1016 : static void
1017 : repair_cmd_fn_metrics( args_t * args,
1018 0 : config_t * config ) {
1019 : //memset( &config->topo, 0, sizeof(config->topo) );
1020 :
1021 0 : fd_topo_join_workspaces( &config->topo, FD_SHMEM_JOIN_MODE_READ_ONLY, FD_TOPO_CORE_DUMP_LEVEL_DISABLED );
1022 0 : fd_topo_fill( &config->topo );
1023 :
1024 0 : ctx_t * repair_ctx;
1025 0 : fd_topo_wksp_t * repair_wksp;
1026 0 : repair_ctx_wksp( args, config, &repair_ctx, &repair_wksp );
1027 :
1028 0 : ulong shred_tile_idx = fd_topo_find_tile( &config->topo, "shred", 0UL );
1029 0 : ulong repair_tile_idx = fd_topo_find_tile( &config->topo, "repair", 0UL );
1030 0 : FD_TEST( shred_tile_idx != ULONG_MAX );
1031 0 : FD_TEST( repair_tile_idx!= ULONG_MAX );
1032 0 : fd_topo_tile_t * shred_tile = &config->topo.tiles[ shred_tile_idx ];
1033 0 : fd_topo_tile_t * repair_tile = &config->topo.tiles[ repair_tile_idx ];
1034 :
1035 0 : volatile ulong * shred_metrics = fd_metrics_tile( shred_tile->metrics );
1036 0 : FD_TEST( shred_metrics );
1037 :
1038 0 : volatile ulong * repair_metrics = fd_metrics_tile( repair_tile->metrics );
1039 0 : FD_TEST( repair_metrics );
1040 0 : ulong * repair_metrics_prev = aligned_alloc( 8UL, sizeof(ulong) * FD_METRICS_TOTAL_SZ );
1041 0 : FD_TEST( repair_metrics_prev );
1042 0 : memset( repair_metrics_prev, 0, sizeof(ulong) * FD_METRICS_TOTAL_SZ );
1043 :
1044 :
1045 0 : ulong net_tile_cnt = config->layout.net_tile_count;
1046 0 : volatile ulong ** repair_net_links = aligned_alloc( 8UL, net_tile_cnt * sizeof(volatile ulong*) );
1047 0 : volatile ulong ** net_shred_links = aligned_alloc( 8UL, net_tile_cnt * sizeof(volatile ulong*) );
1048 0 : FD_TEST( repair_net_links );
1049 0 : FD_TEST( net_shred_links );
1050 :
1051 0 : for( ulong i = 0UL; i < net_tile_cnt; i++ ) {
1052 : /* process all repair_net links */
1053 0 : ulong tile_idx = fd_topo_find_tile( &config->topo, "net", i );
1054 0 : if( FD_UNLIKELY( tile_idx == ULONG_MAX ) ) FD_LOG_ERR(( "net tile %lu not found", i ));
1055 0 : fd_topo_tile_t * tile = &config->topo.tiles[ tile_idx ];
1056 :
1057 0 : ulong repair_net_in_idx = fd_topo_find_tile_in_link( &config->topo, tile, "repair_net", 0UL );
1058 0 : if( FD_UNLIKELY( repair_net_in_idx == ULONG_MAX ) ) FD_LOG_ERR(( "repair_net link not found for net tile %lu", i ));
1059 0 : repair_net_links[i] = fd_metrics_link_in( tile->metrics, repair_net_in_idx );
1060 0 : FD_TEST( repair_net_links[i] );
1061 :
1062 : /* process all net_shred links */
1063 0 : tile_idx = fd_topo_find_tile( &config->topo, "shred", 0 );
1064 0 : if( FD_UNLIKELY( tile_idx == ULONG_MAX ) ) FD_LOG_ERR(( "shred tile 0 not found" ));
1065 0 : fd_topo_tile_t * shred_tile = &config->topo.tiles[ tile_idx ];
1066 :
1067 0 : ulong shred_out_in_idx = fd_topo_find_tile_in_link( &config->topo, shred_tile, "net_shred", i );
1068 0 : if( FD_UNLIKELY( shred_out_in_idx == ULONG_MAX ) ) FD_LOG_ERR(( "net_shred link not found for shred tile 0" ));
1069 0 : net_shred_links[i] = fd_metrics_link_in( shred_tile->metrics, shred_out_in_idx );
1070 0 : FD_TEST( net_shred_links[i] );
1071 0 : }
1072 :
1073 0 : long last_print_ts = fd_log_wallclock();
1074 0 : ulong last_sent = 0UL;
1075 0 : for(;;) {
1076 0 : long now = fd_log_wallclock();
1077 0 : if( FD_UNLIKELY( now - last_print_ts > 1e9L ) ) {
1078 0 : print_tile_metrics( shred_metrics, repair_metrics, repair_metrics_prev, repair_net_links, net_shred_links, net_tile_cnt, &last_sent, last_print_ts, now );
1079 0 : last_print_ts = now;
1080 0 : }
1081 0 : }
1082 0 : }
1083 :
1084 : static void
1085 : repair_cmd_fn_forest( args_t * args,
1086 0 : config_t * config ) {
1087 0 : ctx_t * repair_ctx;
1088 0 : fd_topo_wksp_t * repair_wksp;
1089 0 : repair_ctx_wksp( args, config, &repair_ctx, &repair_wksp );
1090 :
1091 0 : ulong forest_gaddr = fd_wksp_gaddr_fast( repair_ctx->wksp, repair_ctx->forest );
1092 0 : fd_forest_t * forest = (fd_forest_t *)fd_wksp_laddr( repair_wksp->wksp, forest_gaddr );
1093 :
1094 0 : for( ;; ) {
1095 0 : fd_forest_print( forest );
1096 0 : sleep( 1 );
1097 0 : }
1098 0 : }
1099 :
1100 : static void
1101 : repair_cmd_fn_inflight( args_t * args,
1102 0 : config_t * config ) {
1103 0 : ctx_t * repair_ctx;
1104 0 : fd_topo_wksp_t * repair_wksp;
1105 0 : repair_ctx_wksp( args, config, &repair_ctx, &repair_wksp );
1106 :
1107 0 : ulong inflights_gaddr = fd_wksp_gaddr_fast( repair_ctx->wksp, repair_ctx->inflights );
1108 0 : fd_inflights_t * inflights = (fd_inflights_t *)fd_wksp_laddr( repair_wksp->wksp, inflights_gaddr );
1109 :
1110 0 : ulong inflight_pool_off = (ulong)inflights->pool - (ulong)repair_ctx->inflights;
1111 0 : fd_inflight_t * inflight_pool = (fd_inflight_t *)fd_wksp_laddr( repair_wksp->wksp, inflights_gaddr + inflight_pool_off );
1112 :
1113 0 : for( ;; ) {
1114 0 : fd_inflights_print( inflights->outstanding_dl, inflight_pool );
1115 0 : printf("popped count: %lu\n", inflights->popped_cnt);
1116 0 : fd_inflights_print( inflights->popped_dl, inflight_pool );
1117 0 : sleep( 1 );
1118 0 : }
1119 0 : }
1120 :
1121 : static void
1122 : repair_cmd_fn_requests( args_t * args,
1123 0 : config_t * config ) {
1124 0 : ctx_t * repair_ctx;
1125 0 : fd_topo_wksp_t * repair_wksp;
1126 0 : repair_ctx_wksp( args, config, &repair_ctx, &repair_wksp );
1127 :
1128 0 : fd_forest_t * forest = fd_forest_join( fd_wksp_laddr( repair_wksp->wksp, fd_wksp_gaddr_fast( repair_ctx->wksp, repair_ctx->forest ) ) );
1129 0 : fd_forest_reqslist_t * dlist = fd_forest_reqslist( forest );
1130 0 : fd_forest_ref_t * pool = fd_forest_reqspool( forest );
1131 :
1132 0 : fd_forest_reqslist_t * orphlist = fd_forest_orphlist( forest );
1133 :
1134 0 : for( ;; ) {
1135 0 : printf("%-15s %-12s %-12s %-12s %-20s %-12s\n",
1136 0 : "Slot", "Buffered Idx", "Complete Idx", "First Shred ts", "Turbine Cnt", "Repair Cnt");
1137 0 : printf("%-15s %-12s %-12s %-12s %-20s %-12s\n",
1138 0 : "---------------", "------------", "------------", "------------",
1139 0 : "--------------------", "------------");
1140 0 : for( fd_forest_reqslist_iter_t iter = fd_forest_reqslist_iter_fwd_init( dlist, pool );
1141 0 : !fd_forest_reqslist_iter_done( iter, dlist, pool );
1142 0 : iter = fd_forest_reqslist_iter_fwd_next( iter, dlist, pool ) ) {
1143 0 : fd_forest_ref_t * req = fd_forest_reqslist_iter_ele( iter, dlist, pool );
1144 0 : fd_forest_blk_t * blk = fd_forest_pool_ele( fd_forest_pool( forest ), req->idx );
1145 :
1146 0 : printf("%-15lu %-12u %-12u %-20ld %-12u %-10u\n",
1147 0 : blk->slot,
1148 0 : blk->buffered_idx,
1149 0 : blk->complete_idx,
1150 0 : blk->first_shred_ts,
1151 0 : blk->turbine_cnt,
1152 0 : blk->repair_cnt);
1153 0 : }
1154 0 : printf("\n");
1155 :
1156 : /* now lets print the orphreqs */
1157 :
1158 0 : printf("Orphan Requests:\n");
1159 0 : printf("%-15s %-12s %-12s %-12s %-20s %-12s %-10s\n",
1160 0 : "Slot", "Consumed Idx", "Buffered Idx", "Complete Idx",
1161 0 : "First Shred Timestamp", "Turbine Cnt", "Repair Cnt");
1162 0 : printf("%-15s %-12s %-12s %-12s %-20s %-12s %-10s\n",
1163 0 : "---------------", "------------", "------------", "------------",
1164 0 : "--------------------", "------------", "----------");
1165 :
1166 0 : for( fd_forest_reqslist_iter_t iter = fd_forest_reqslist_iter_fwd_init( orphlist, pool );
1167 0 : !fd_forest_reqslist_iter_done( iter, orphlist, pool );
1168 0 : iter = fd_forest_reqslist_iter_fwd_next( iter, orphlist, pool ) ) {
1169 0 : fd_forest_ref_t * req = fd_forest_reqslist_iter_ele( iter, orphlist, pool );
1170 0 : fd_forest_blk_t * blk = fd_forest_pool_ele( fd_forest_pool( forest ), req->idx );
1171 0 : printf("%-15lu %-12u %-12u %-20ld %-12u %-10u\n",
1172 0 : blk->slot,
1173 0 : blk->buffered_idx,
1174 0 : blk->complete_idx,
1175 0 : blk->first_shred_ts,
1176 0 : blk->turbine_cnt,
1177 0 : blk->repair_cnt);
1178 0 : }
1179 0 : sleep( 1 );
1180 0 : }
1181 0 : }
1182 :
1183 : static void
1184 : repair_cmd_fn_waterfall( args_t * args,
1185 0 : config_t * config ) {
1186 :
1187 0 : fd_topo_t * topo = &config->topo;
1188 0 : ulong wksp_id = fd_topo_find_wksp( topo, "repair" );
1189 0 : if( FD_UNLIKELY( wksp_id==ULONG_MAX ) ) FD_LOG_ERR(( "repair workspace not found" ));
1190 0 : fd_topo_wksp_t * repair_wksp = &topo->workspaces[ wksp_id ];
1191 0 : fd_topo_join_workspace( topo, repair_wksp, FD_SHMEM_JOIN_MODE_READ_ONLY, FD_TOPO_CORE_DUMP_LEVEL_DISABLED );
1192 :
1193 : /* Access the repair tile scratch memory where repair_tile_ctx is stored */
1194 0 : ulong tile_id = fd_topo_find_tile( topo, "repair", 0UL );
1195 0 : if( FD_UNLIKELY( tile_id==ULONG_MAX ) ) FD_LOG_ERR(( "repair tile not found" ));
1196 0 : fd_topo_tile_t * tile = &topo->tiles[ tile_id ];
1197 0 : void * scratch = fd_topo_obj_laddr( &config->topo, tile->tile_obj_id );
1198 0 : if( FD_UNLIKELY( !scratch ) ) FD_LOG_ERR(( "Failed to access repair tile scratch memory" ));
1199 :
1200 0 : FD_SCRATCH_ALLOC_INIT( l, scratch );
1201 0 : ctx_t * repair_ctx = FD_SCRATCH_ALLOC_APPEND( l, alignof(ctx_t), sizeof(ctx_t) );
1202 :
1203 : /* catchup cmd owned memory */
1204 0 : location_table = fd_location_table_join( fd_location_table_new( location_table_mem ) );
1205 0 : read_iptable( args->repair.iptable_path, location_table );
1206 :
1207 : // Add terminal setup here - same as monitor.c
1208 0 : atexit( restore_terminal );
1209 0 : if( FD_UNLIKELY( 0!=tcgetattr( STDIN_FILENO, &termios_backup ) ) ) {
1210 0 : FD_LOG_ERR(( "tcgetattr(STDIN_FILENO) failed (%i-%s)", errno, fd_io_strerror( errno ) ));
1211 0 : }
1212 :
1213 : /* Disable character echo and line buffering */
1214 0 : struct termios term = termios_backup;
1215 0 : term.c_lflag &= (tcflag_t)~(ICANON | ECHO);
1216 0 : if( FD_UNLIKELY( 0!=tcsetattr( STDIN_FILENO, TCSANOW, &term ) ) ) {
1217 0 : FD_LOG_WARNING(( "tcsetattr(STDIN_FILENO) failed (%i-%s)", errno, fd_io_strerror( errno ) ));
1218 0 : }
1219 :
1220 0 : int catchup_verbose = 0;
1221 0 : long last_print = 0;
1222 0 : for( ;; ) {
1223 0 : int c = fd_getchar();
1224 0 : if( FD_UNLIKELY( c=='i' ) ) catchup_verbose = !catchup_verbose;
1225 0 : if( FD_UNLIKELY( c=='\x04' ) ) break; /* Ctrl-D */
1226 :
1227 0 : long now = fd_log_wallclock();
1228 0 : if( FD_UNLIKELY( now - last_print > 1e9L ) ) {
1229 0 : last_print = now;
1230 0 : print_catchup_slots( repair_wksp->wksp, repair_ctx, catchup_verbose, args->repair.sort_by_slot );
1231 0 : printf( "catchup slots | Use 'i' to toggle extra slot information" TEXT_NEWLINE );
1232 0 : fflush( stdout );
1233 :
1234 : /* Peer location latency is not that useful post catchup, and also
1235 : requires some concurrent dlist iteration, so only print it when
1236 : in catchup mode. */
1237 0 : }
1238 0 : }
1239 0 : }
1240 :
1241 0 : #define PEERS_DISPLAY_MAX 20
1242 :
1243 : static void
1244 : print_peer_dlist( fd_policy_peer_dlist_t * dlist,
1245 : fd_policy_peer_t * pool,
1246 : fd_policy_peer_dlist_iter_t cursor,
1247 0 : char const * label ) {
1248 0 : ulong cnt = 0;
1249 0 : for( fd_policy_peer_dlist_iter_t it = fd_policy_peer_dlist_iter_fwd_init( dlist, pool );
1250 0 : !fd_policy_peer_dlist_iter_done( it, dlist, pool );
1251 0 : it = fd_policy_peer_dlist_iter_fwd_next( it, dlist, pool ) ) cnt++;
1252 :
1253 0 : printf( "%s (%lu peers)\n", label, cnt );
1254 0 : if( !cnt || fd_policy_peer_dlist_iter_done( cursor, dlist, pool ) ) {
1255 0 : printf( " (empty or iterator not initialized)\n\n" );
1256 0 : return;
1257 0 : }
1258 :
1259 0 : printf( " | %-8s | %-12s | %-12s | %-8s | %-8s\n",
1260 0 : "Idx", "Pubkey", "Ewma Lat", "Avg Lat", "Req/Res" );
1261 0 : printf( "-----+----------+--------------+--------------+----------+---------\n" );
1262 :
1263 0 : fd_policy_peer_dlist_iter_t it = cursor;
1264 0 : for( ulong i = 0; i < PEERS_DISPLAY_MAX && i < cnt; i++ ) {
1265 0 : fd_policy_peer_t * peer = fd_policy_peer_dlist_iter_ele( it, dlist, pool );
1266 :
1267 0 : FD_BASE58_ENCODE_32_BYTES( peer->key.key, b58 );
1268 0 : char pubkey_short[13];
1269 0 : fd_cstr_fini( fd_cstr_append_text( fd_cstr_init( pubkey_short ), b58, 12 ) );
1270 :
1271 0 : double avg_lat_ms = peer->res_cnt ? ((double)peer->total_lat / (double)peer->res_cnt) / 1e6 : 0.0;
1272 0 : double ewma_lat_ms = (double)peer->ewma_lat / 1e6;
1273 :
1274 0 : printf( " %s%c%s | %-8lu | %-12s | %9.3fms | %9.3fms | %lu/%lu\n",
1275 0 : i == 0 ? "\033[1;33m" : "",
1276 0 : i == 0 ? '>' : ' ',
1277 0 : i == 0 ? "\033[0m" : "",
1278 0 : fd_policy_peer_pool_idx( pool, peer ),
1279 0 : pubkey_short,
1280 0 : ewma_lat_ms,
1281 0 : avg_lat_ms,
1282 0 : peer->req_cnt,
1283 0 : peer->res_cnt );
1284 :
1285 0 : it = fd_policy_peer_dlist_iter_fwd_next( it, dlist, pool );
1286 0 : if( fd_policy_peer_dlist_iter_done( it, dlist, pool ) ) {
1287 0 : it = fd_policy_peer_dlist_iter_fwd_init( dlist, pool );
1288 0 : }
1289 0 : }
1290 0 : if( cnt > PEERS_DISPLAY_MAX ) printf( " ... (%lu more)\n", cnt - PEERS_DISPLAY_MAX );
1291 0 : printf( "\n" );
1292 0 : }
1293 :
1294 : static void
1295 : repair_cmd_fn_peers( args_t * args,
1296 0 : config_t * config ) {
1297 0 : ctx_t * repair_ctx;
1298 0 : fd_topo_wksp_t * repair_wksp;
1299 0 : repair_ctx_wksp( args, config, &repair_ctx, &repair_wksp );
1300 :
1301 0 : fd_policy_t * policy = fd_wksp_laddr( repair_wksp->wksp, fd_wksp_gaddr_fast( repair_ctx->wksp, repair_ctx->policy ) );
1302 :
1303 0 : fd_policy_peer_dlist_t * fast_dlist = fd_wksp_laddr( repair_wksp->wksp, fd_wksp_gaddr_fast( repair_ctx->wksp, policy->peers.fast ) );
1304 0 : fd_policy_peer_dlist_t * slow_dlist = fd_wksp_laddr( repair_wksp->wksp, fd_wksp_gaddr_fast( repair_ctx->wksp, policy->peers.slow ) );
1305 0 : fd_policy_peer_t * pool = fd_wksp_laddr( repair_wksp->wksp, fd_wksp_gaddr_fast( repair_ctx->wksp, policy->peers.pool ) );
1306 :
1307 0 : long last_print = 0;
1308 0 : for( ;; ) {
1309 0 : long now = fd_log_wallclock();
1310 0 : if( FD_UNLIKELY( now - last_print > 1e9L ) ) {
1311 0 : last_print = now;
1312 0 : printf( "\033[2J\033[H" );
1313 :
1314 0 : char fast_label[64];
1315 0 : char slow_label[64];
1316 0 : snprintf( fast_label, sizeof(fast_label), "FAST PEERS (ewma < %ldms)", (long)(FD_POLICY_LATENCY_THRESH / 1e6L) );
1317 0 : snprintf( slow_label, sizeof(slow_label), "SLOW PEERS (ewma >= %ldms or no responses)", (long)(FD_POLICY_LATENCY_THRESH / 1e6L) );
1318 0 : print_peer_dlist( fast_dlist, pool, policy->peers.select.fast_iter, fast_label );
1319 0 : print_peer_dlist( slow_dlist, pool, policy->peers.select.slow_iter, slow_label );
1320 :
1321 0 : printf( "select cnt: %u / %u (fast per slow)\n", policy->peers.select.cnt, FD_POLICY_FAST_PER_SLOW );
1322 0 : printf( "pool used: %lu / %lu\n", fd_policy_peer_pool_used( pool ), fd_policy_peer_pool_max( pool ) );
1323 :
1324 0 : fflush( stdout );
1325 0 : }
1326 :
1327 0 : }
1328 0 : }
1329 :
1330 :
1331 :
1332 :
1333 : void
1334 : repair_cmd_args( int * pargc,
1335 : char *** pargv,
1336 0 : args_t * args ) {
1337 :
1338 : /* positional arg */
1339 :
1340 0 : args->repair.pos_arg = (*pargv)[0];
1341 0 : if( FD_UNLIKELY( !args->repair.pos_arg ) ) {
1342 0 : args->repair.help = 1;
1343 0 : return;
1344 0 : }
1345 0 : (*pargc)--;
1346 0 : (*pargv)++;
1347 :
1348 : /* required args */
1349 :
1350 0 : char const * manifest_path = fd_env_strip_cmdline_cstr ( pargc, pargv, "--manifest-path", NULL, NULL );
1351 :
1352 : /* optional args */
1353 :
1354 0 : char const * iptable_path = fd_env_strip_cmdline_cstr ( pargc, pargv, "--iptable", NULL, NULL );
1355 0 : ulong slot = fd_env_strip_cmdline_ulong ( pargc, pargv, "--slot", NULL, ULONG_MAX );
1356 0 : int sort_by_slot = fd_env_strip_cmdline_contains( pargc, pargv, "--sort-by-slot" );
1357 :
1358 0 : if( FD_UNLIKELY( !strcmp( args->repair.pos_arg, "catchup" ) && !manifest_path ) ) {
1359 0 : args->repair.help = 1;
1360 0 : return;
1361 0 : }
1362 :
1363 0 : fd_cstr_fini( fd_cstr_append_cstr_safe( fd_cstr_init( args->repair.manifest_path ), manifest_path, sizeof(args->repair.manifest_path)-1UL ) );
1364 0 : fd_cstr_fini( fd_cstr_append_cstr_safe( fd_cstr_init( args->repair.iptable_path ), iptable_path, sizeof(args->repair.iptable_path )-1UL ) );
1365 0 : args->repair.slot = slot;
1366 0 : args->repair.sort_by_slot = sort_by_slot;
1367 0 : }
1368 :
1369 : static void
1370 : repair_cmd_fn( args_t * args,
1371 0 : config_t * config ) {
1372 :
1373 0 : if( args->repair.help ) {
1374 0 : fd_action_help_print( &fd_action_repair );
1375 0 : return;
1376 0 : }
1377 :
1378 0 : if ( !strcmp( args->repair.pos_arg, "catchup" ) ) repair_cmd_fn_catchup ( args, config );
1379 0 : else if( !strcmp( args->repair.pos_arg, "eqvoc" ) ) repair_cmd_fn_eqvoc ( args, config );
1380 0 : else if( !strcmp( args->repair.pos_arg, "forest" ) ) repair_cmd_fn_forest ( args, config );
1381 0 : else if( !strcmp( args->repair.pos_arg, "inflight" ) ) repair_cmd_fn_inflight ( args, config );
1382 0 : else if( !strcmp( args->repair.pos_arg, "requests" ) ) repair_cmd_fn_requests ( args, config );
1383 0 : else if( !strcmp( args->repair.pos_arg, "waterfall" ) ) repair_cmd_fn_waterfall( args, config );
1384 0 : else if( !strcmp( args->repair.pos_arg, "peers" ) ) repair_cmd_fn_peers ( args, config );
1385 0 : else if( !strcmp( args->repair.pos_arg, "metrics" ) ) repair_cmd_fn_metrics ( args, config );
1386 0 : else fd_action_help_print( &fd_action_repair );
1387 0 : }
1388 :
1389 : static void
1390 0 : repair_args_help( fd_action_help_t * help ) {
1391 0 : fd_action_help_arg( help, "catchup", NULL, "Run a reduced topology that only repairs slots until catchup.\n"
1392 0 : "Requires --manifest-path; accepts --iptable and --sort-by-slot" );
1393 0 : fd_action_help_arg( help, "eqvoc", NULL, "Test equivocation detection and the repair path" );
1394 0 : fd_action_help_arg( help, "forest", NULL, "Print the repair forest. Accepts --slot to drill into a slot" );
1395 0 : fd_action_help_arg( help, "inflight", NULL, "Print the inflight repairs" );
1396 0 : fd_action_help_arg( help, "requests", NULL, "Print the queued repair requests" );
1397 0 : fd_action_help_arg( help, "waterfall", NULL, "Print a waterfall diagram of recent slot completion times and\n"
1398 0 : "response latencies. Accepts --iptable and --sort-by-slot" );
1399 0 : fd_action_help_arg( help, "peers", NULL, "Print the list of slow and fast repair peers" );
1400 0 : fd_action_help_arg( help, "metrics", NULL, "Print repair tile metrics in a digestible format" );
1401 0 : fd_action_help_arg( help, "--manifest-path", "<path>", "Path to manifest file (required by catchup)" );
1402 0 : fd_action_help_arg( help, "--iptable", "<path>", "Path to iptable file (catchup, waterfall)" );
1403 0 : fd_action_help_arg( help, "--slot", "<slot>", "Specific forest slot to drill into (forest)" );
1404 : fd_action_help_arg( help, "--sort-by-slot", NULL, "Sort results by slot (catchup, waterfall)" );
1405 0 : }
1406 :
1407 : action_t fd_action_repair = {
1408 : .name = "repair",
1409 : .args = repair_cmd_args,
1410 : .fn = repair_cmd_fn,
1411 : .perm = dev_cmd_perm,
1412 : .description = "Spawn a reduced topology for inspecting and profiling the repair tile",
1413 : .detail = "Boots a smaller Firedancer topology focused on the repair tile and runs the\n"
1414 : "requested subcommand to drive or inspect repair behavior. Pick one of the\n"
1415 : "subcommands below.",
1416 : .usage = "repair <catchup|eqvoc|forest|inflight|requests|waterfall|peers|metrics> [OPTIONS]",
1417 : .args_help = repair_args_help,
1418 : };
|
