LU-14090 mgs: no local logs flag

There is a feature that starts a target with a local copy of
config log in order to avoid a delay in communicating with
an MGS and to load mgs log updates later on. However, that
feature is not always useful.

When replace_nids adds records with new nids it does not
append remote config logs but overwrite corresponding
records in place. If a target starts using local config
log - it gets confused by outdated nids.

This patch adds tunefs.lustre --nolocallogs key that
sets nolocallogs flag, which says ignore local configs copy.
The flag is reset once new logs are uploaded from MGS.

tunefs.lustre --nolocallogs is suggested to be executed on
targets together with replace_nids on MGS.

HPE-bug-id: LUS-2510
Change-Id: I949c19ac701d287e1c1199bc12445989476a707b
Signed-off-by: Artem Blagodarenko <artem.blagodarenko@hpe.com>
Reviewed-on: https://es-gerrit.dev.cray.com/157574
Reviewed-by: Vladimir Saveliev <c17830@cray.com>
Reviewed-by: Nikitas Angelinas <nangelinas@cray.com>
Tested-by: Alexander Lezhoev <c17454@cray.com>
Reviewed-on: https://review.whamcloud.com/40448
Tested-by: jenkins <devops@whamcloud.com>
Reviewed-by: Andreas Dilger <adilger@whamcloud.com>
Tested-by: Maloo <maloo@whamcloud.com>
Reviewed-by: Oleg Drokin <green@whamcloud.com>

diff --git a/lustre/doc/tunefs.lustre.8 b/lustre/doc/tunefs.lustre.8
index 3d8c6b6..bde1071 100644 (file)
--- a/lustre/doc/tunefs.lustre.8
+++ b/lustre/doc/tunefs.lustre.8
@@ -134,6 +134,19 @@ Correct order of operations is:
 .br
 * Mount clients
 .TP
+.BI \--nolocallogs
+Use configuration logs from the MGS, not local ones.
+.br
+There is feature that starts a target with a local copy of
+the config log in order to avoid a delay in communicating with
+the MGS and to load MGS log updates later on.
+However, that feature is not always useful.
+.br
+replace_nids changes config logs on the server side and local copies
+become invalid. --nolocallogs sets the nolocallogs mount flag,
+which causes the local copy of the config log to be ignored.
+The flag is reset once new logs are downloaded from MGS.
+.TP
 .BI \--quota
 Enable space accounting on old 2.x devices.
 

diff --git a/lustre/include/lustre_disk.h b/lustre/include/lustre_disk.h
index 3a071d4..4821985 100644 (file)
--- a/lustre/include/lustre_disk.h
+++ b/lustre/include/lustre_disk.h
@@ -109,6 +109,7 @@ struct lustre_mount_data {
 #define LMD_FLG_NO_PRECREATE   0x10000 /* do not allow OST object creation */
 #define LMD_FLG_LOCAL_RECOV    0x20000 /* force recovery for local clients */
 #define LMD_FLG_ABORT_RECOV_MDT        0x40000 /* Abort recovery between MDTs */
+#define LMD_FLG_NO_LOCAL_LOGS  0x80000 /* Use config logs from MGS */
 
 #define lmd_is_client(x) ((x)->lmd_flags & LMD_FLG_CLIENT)
 

diff --git a/lustre/include/uapi/linux/lustre/lustre_disk.h b/lustre/include/uapi/linux/lustre/lustre_disk.h
index e9cbf30..458eb3a 100644 (file)
--- a/lustre/include/uapi/linux/lustre/lustre_disk.h
+++ b/lustre/include/uapi/linux/lustre/lustre_disk.h
@@ -102,6 +102,8 @@
 #define LDD_F_ERROR            0x4000
 /** process at lctl conf_param */
 #define LDD_F_PARAM2           0x8000
+/** the target shouldn't use local logs */
+#define LDD_F_NO_LOCAL_LOGS    0x10000
 
 #define LDD_MAGIC 0x1dd00001
 

diff --git a/lustre/mgc/mgc_request.c b/lustre/mgc/mgc_request.c
index 0d06d75..ecce505 100644 (file)
--- a/lustre/mgc/mgc_request.c
+++ b/lustre/mgc/mgc_request.c
@@ -1924,21 +1924,32 @@ static int mgc_process_cfg_log(struct obd_device *mgc,
            cli->cl_mgc_configs_dir != NULL &&
            lu2dt_dev(cli->cl_mgc_configs_dir->do_lu.lo_dev) ==
            lsi->lsi_dt_dev) {
-               if (!local_only && !lsi->lsi_dt_dev->dd_rdonly)
+               if (!local_only && !lsi->lsi_dt_dev->dd_rdonly) {
                        /* Only try to copy log if we have the lock. */
+                       CDEBUG(D_INFO, "%s: copy local log %s\n",
+                              mgc->obd_name, cld->cld_logname);
+
                        rc = mgc_llog_local_copy(env, mgc, ctxt, lctxt,
                                                 cld->cld_logname);
+                       if (!rc)
+                               lsi->lsi_flags &= ~LDD_F_NO_LOCAL_LOGS;
+               }
                if (local_only || rc) {
+                       if (unlikely(lsi->lsi_flags & LDD_F_NO_LOCAL_LOGS)
+                           || rc) {
+                               CWARN("%s: local log %s are not valid and/or remote logs are not accessbile rc = %d\n",
+                                     mgc->obd_name, cld->cld_logname, rc);
+                               GOTO(out_pop, rc = -EIO);
+                       }
+
                        if (strcmp(cld->cld_logname, PARAMS_FILENAME) != 0 &&
                            llog_is_empty(env, lctxt, cld->cld_logname)) {
-                               LCONSOLE_ERROR_MSG(0x13a, "Failed to get MGS "
-                                                  "log %s and no local copy."
-                                                  "\n", cld->cld_logname);
+                               LCONSOLE_ERROR_MSG(0x13a, "Failed to get MGS log %s and no local copy.\n",
+                                                  cld->cld_logname);
                                GOTO(out_pop, rc = -ENOENT);
                        }
-                       CDEBUG(D_MGC, "Failed to get MGS log %s, using local "
-                              "copy for now, will try to update later.\n",
-                              cld->cld_logname);
+                       CDEBUG(D_MGC, "%s: Failed to get MGS log %s, using local copy for now, will try to update later.\n",
+                              mgc->obd_name, cld->cld_logname);
                        rc = 0;
                }
                /* Now, whether we copied or not, start using the local llog.

diff --git a/lustre/obdclass/obd_mount.c b/lustre/obdclass/obd_mount.c
index 37165a2..be1521f 100644 (file)
--- a/lustre/obdclass/obd_mount.c
+++ b/lustre/obdclass/obd_mount.c
@@ -1395,6 +1395,9 @@ int lmd_parse(char *options, struct lustre_mount_data *lmd)
                } else if (strncmp(s1, "writeconf", 9) == 0) {
                        lmd->lmd_flags |= LMD_FLG_WRITECONF;
                        clear++;
+               } else if (strncmp(s1, "nolocallogs", 11) == 0) {
+                       lmd->lmd_flags |= LMD_FLG_NO_LOCAL_LOGS;
+                       clear++;
                } else if (strncmp(s1, "update", 6) == 0) {
                        lmd->lmd_flags |= LMD_FLG_UPDATE;
                        clear++;

diff --git a/lustre/obdclass/obd_mount_server.c b/lustre/obdclass/obd_mount_server.c
index acd9222..374ec8b 100644 (file)
--- a/lustre/obdclass/obd_mount_server.c
+++ b/lustre/obdclass/obd_mount_server.c
@@ -1553,6 +1553,8 @@ static int lsi_prepare(struct lustre_sb_info *lsi)
         */
        lsi->lsi_flags |= (lsi->lsi_lmd->lmd_flags & LMD_FLG_WRITECONF) ?
                LDD_F_WRITECONF : 0;
+       lsi->lsi_flags |= (lsi->lsi_lmd->lmd_flags & LMD_FLG_NO_LOCAL_LOGS) ?
+               LDD_F_NO_LOCAL_LOGS : 0;
        lsi->lsi_flags |= (lsi->lsi_lmd->lmd_flags & LMD_FLG_VIRGIN) ?
                LDD_F_VIRGIN : 0;
        lsi->lsi_flags |= (lsi->lsi_lmd->lmd_flags & LMD_FLG_UPDATE) ?

diff --git a/lustre/tests/conf-sanity.sh b/lustre/tests/conf-sanity.sh
index 2e390d1..f540da2 100644 (file)
--- a/lustre/tests/conf-sanity.sh
+++ b/lustre/tests/conf-sanity.sh
@@ -9125,6 +9125,43 @@ test_127() {
 }
 run_test 127 "direct io overwrite on full ost"
 
+test_128()
+{
+       combined_mgs_mds && skip "need separate mgs device"
+       [ "$ost2_FSTYPE" == zfs ] && import_zpool ost2
+
+       format_ost 2
+       # Try to apply nolocallogs to the virgin OST. Should fail.
+       do_facet ost2 "$TUNEFS --nolocallogs $(ostdevname 2)" &&
+               error "nolocallogs should not be allowed on the virgin target"
+
+       setupall
+       stopall
+
+       [ "$ost1_FSTYPE" == zfs ] && import_zpool ost1
+       # Start OST without MGS (local configs)
+       do_facet ost1 "$TUNEFS --dryrun $(ostdevname 1)"
+       start_ost || error "unable to start OST1"
+       stop_ost || error "Unable to stop OST1"
+
+       [ "$ost1_FSTYPE" == zfs ] && import_zpool ost1
+       # Do not allow reading local configs, should fail
+       do_facet ost1 "$TUNEFS --nolocallogs $(ostdevname 1)" ||
+               error "Can not set nolocallogs"
+       start_ost && error "OST1 started, but should fail"
+
+       # Connect to MGS successfully, reset nolocallogs flag
+       [ "$ost1_FSTYPE" == zfs ] && import_zpool ost1
+       start_mgs || error "unable to start MGS"
+       start_ost || error "unable to start OST1"
+
+       do_facet ost1 "$TUNEFS --dryrun $(ostdevname 1)" | grep "nolocallogs" &&
+               error "nolocallogs expected to be reset"
+
+       stop_ost || error "Unable to stop OST1"
+}
+run_test 128 "Force using remote logs with --nolocallogs"
+
 if ! combined_mgs_mds ; then
        stop mgs
 fi

diff --git a/lustre/utils/mkfs_lustre.c b/lustre/utils/mkfs_lustre.c
index a366039..b1f0f58 100644 (file)
--- a/lustre/utils/mkfs_lustre.c
+++ b/lustre/utils/mkfs_lustre.c
@@ -156,6 +156,7 @@ void usage(FILE *out)
                "\t\t--erase-param <key>: erase all instances of a parameter\n"
                "\t\t--erase-params: erase all old parameter settings\n"
                "\t\t--writeconf: erase all config logs for this fs.\n"
+               "\t\t--nolocallogs: use logs from MGS, not local ones.\n"
                "\t\t--quota: enable space accounting on old 2.x device.\n"
                "\t\t--rename: rename the filesystem name\n"
 #endif
@@ -187,7 +188,7 @@ void print_ldd(char *str, struct mkfs_opts *mop)
        printf("Lustre FS:  %s\n", ldd->ldd_fsname);
        printf("Mount type: %s\n", MT_STR(ldd));
        printf("Flags:      %#x\n", ldd->ldd_flags);
-       printf("              (%s%s%s%s%s%s%s%s)\n",
+       printf("              (%s%s%s%s%s%s%s%s%s)\n",
               IS_MDT(ldd) ? "MDT " : "",
               IS_OST(ldd) ? "OST " : "",
               IS_MGS(ldd) ? "MGS " : "",
@@ -195,7 +196,8 @@ void print_ldd(char *str, struct mkfs_opts *mop)
               ldd->ldd_flags & LDD_F_VIRGIN     ? "first_time " : "",
               ldd->ldd_flags & LDD_F_UPDATE     ? "update " : "",
               ldd->ldd_flags & LDD_F_WRITECONF  ? "writeconf " : "",
-              ldd->ldd_flags & LDD_F_NO_PRIMNODE ? "no_primnode " : "");
+              ldd->ldd_flags & LDD_F_NO_PRIMNODE ? "no_primnode " : "",
+              ldd->ldd_flags & LDD_F_NO_LOCAL_LOGS ? "nolocallogs " : "");
        printf("Persistent mount opts: %s\n", ldd->ldd_mount_opts);
        osd_print_ldd_params(mop);
        if (ldd->ldd_userdata[0])
@@ -228,10 +230,11 @@ static bool server_make_name(__u32 flags, __u16 index, const char *fs,
        if (flags & (LDD_F_SV_TYPE_MDT | LDD_F_SV_TYPE_OST)) {
                if (!(flags & LDD_F_SV_ALL))
                        snprintf(name_buf, name_buf_size, "%.8s%c%s%04x", fs,
-                                (flags & LDD_F_VIRGIN) ? ':' :
-                                ((flags & LDD_F_WRITECONF) ? '=' : '-'),
-                                (flags & LDD_F_SV_TYPE_MDT) ? "MDT" : "OST",
-                                 index);
+                               (flags & LDD_F_VIRGIN) ? ':' :
+                               ((flags & LDD_F_WRITECONF) ? '=' :
+                               ((flags & LDD_F_NO_LOCAL_LOGS) ? '+' : '-')),
+                               (flags & LDD_F_SV_TYPE_MDT) ? "MDT" : "OST",
+                               index);
        } else if (flags & LDD_F_SV_TYPE_MGS) {
                snprintf(name_buf, name_buf_size, "MGS");
        } else {
@@ -399,6 +402,7 @@ int parse_opts(int argc, char *const argv[], struct mkfs_opts *mop,
        { .val = 'E',   .name =  "erase-param", .has_arg = required_argument},
        { .val = 'e',   .name =  "erase-params",
                                                .has_arg = no_argument},
+       { .val = 'l',   .name =  "nolocallogs", .has_arg = no_argument},
        { .val = 'Q',   .name =  "quota",       .has_arg = no_argument},
        { .val = 'R',   .name =  "rename",      .has_arg = optional_argument},
        { .val = 'w',   .name =  "writeconf",   .has_arg = no_argument},
@@ -408,7 +412,7 @@ int parse_opts(int argc, char *const argv[], struct mkfs_opts *mop,
 #ifndef TUNEFS
                          "b:c:d:k:MOrR";
 #else
-                         "E:eQR::w";
+                         "E:elQR::w";
 #endif
        struct lustre_disk_data *ldd = &mop->mo_ldd;
        char new_fsname[16] = { 0 };
@@ -737,6 +741,13 @@ int parse_opts(int argc, char *const argv[], struct mkfs_opts *mop,
                case 'w':
                        ldd->ldd_flags |= LDD_F_WRITECONF;
                        break;
+               case 'l':
+                       if (ldd->ldd_flags & (LDD_F_VIRGIN | LDD_F_WRITECONF)) {
+                               fprintf(stderr, "Can not apply nolocallogs to the target that was writeconfed or never been registered\n");
+                               return EINVAL;
+                       }
+                       ldd->ldd_flags |= LDD_F_NO_LOCAL_LOGS;
+                       break;
 #endif /* !TUNEFS */
                default:
                        if (opt != '?') {
@@ -870,7 +881,8 @@ int main(int argc, char *const argv[])
        }
 
        strscpy(old_fsname, ldd->ldd_fsname, sizeof(ldd->ldd_fsname));
-       ldd->ldd_flags &= ~(LDD_F_WRITECONF | LDD_F_VIRGIN);
+       ldd->ldd_flags &= ~(LDD_F_WRITECONF | LDD_F_VIRGIN |
+                           LDD_F_NO_LOCAL_LOGS);
 
        /* svname of the form lustre:OST1234 means never registered */
        ret = strlen(ldd->ldd_svname);
@@ -880,6 +892,9 @@ int main(int argc, char *const argv[])
        } else if (ldd->ldd_svname[ret - 8] == '=') {
                ldd->ldd_svname[ret - 8] = '-';
                ldd->ldd_flags |= LDD_F_WRITECONF;
+       } else if (ldd->ldd_svname[ret - 8] == '+') {
+               ldd->ldd_svname[ret - 8] = '-';
+               ldd->ldd_flags |= LDD_F_NO_LOCAL_LOGS;
        }
 
        if (strstr(ldd->ldd_params, PARAM_MGSNODE))
@@ -1007,8 +1022,11 @@ int main(int argc, char *const argv[])
                goto out;
        }
 
-       if (check_mtab_entry(mop.mo_device, mop.mo_device, NULL, NULL))
+       if (check_mtab_entry(mop.mo_device, mop.mo_device, NULL, NULL)) {
+               fprintf(stderr, "%s: is currently mounted, exiting without any change\n",
+                       mop.mo_device);
                return EEXIST;
+       }
 
        /* Create the loopback file */
        if (mop.mo_flags & MO_IS_LOOP) {
@@ -1064,6 +1082,15 @@ int main(int argc, char *const argv[])
                (void)osd_label_lustre(&opts);
        }
 
+       /* update svname with '+' to force remote logs */
+       if (ldd->ldd_flags & LDD_F_NO_LOCAL_LOGS) {
+               struct mount_opts opts;
+
+               opts.mo_ldd = *ldd;
+               opts.mo_source = mop.mo_device;
+               (void) osd_label_lustre(&opts);
+       }
+
        /* Rename filesystem fsname */
        if (mop.mo_flags & MO_RENAME) {
                ret = osd_rename_fsname(&mop, old_fsname);

diff --git a/lustre/utils/mount_lustre.c b/lustre/utils/mount_lustre.c
index 1c9c79b..7b363a5 100644 (file)
--- a/lustre/utils/mount_lustre.c
+++ b/lustre/utils/mount_lustre.c
@@ -560,7 +560,8 @@ static int parse_ldd(char *source, struct mount_opts *mop,
                clear_update_ondisk(source, ldd);
 
        /* Since we never rewrite ldd, ignore temp flags */
-       ldd->ldd_flags &= ~(LDD_F_VIRGIN | LDD_F_WRITECONF);
+       ldd->ldd_flags &= ~(LDD_F_VIRGIN | LDD_F_WRITECONF |
+                           LDD_F_NO_LOCAL_LOGS);
 
        /* This is to make sure default options go first */
        temp_options = strdup(options);
@@ -588,6 +589,9 @@ static int parse_ldd(char *source, struct mount_opts *mop,
                } else if (ldd->ldd_svname[rc - 8] == '=') {
                        ldd->ldd_svname[rc - 8] = '-';
                        ldd->ldd_flags |= LDD_F_WRITECONF;
+               } else if (ldd->ldd_svname[rc - 8] == '+') {
+                       ldd->ldd_svname[rc - 8] = '-';
+                       ldd->ldd_flags |= LDD_F_NO_LOCAL_LOGS;
                }
        }
        /* backend osd type */
@@ -634,6 +638,11 @@ static int parse_ldd(char *source, struct mount_opts *mop,
                if (rc != 0)
                        return rc;
        }
+       if (ldd->ldd_flags & LDD_F_NO_LOCAL_LOGS) {
+               rc = append_option(options, options_len, "nolocallogs", NULL);
+               if (rc != 0)
+                       return rc;
+       }
        if (ldd->ldd_flags & LDD_F_NO_PRIMNODE) {
                rc = append_option(options, options_len, "noprimnode", NULL);
                if (rc != 0)
@@ -804,7 +813,8 @@ static void label_lustre(struct mount_opts *mop)
        if (mop->mo_nosvc)
                return;
 
-       if (mop->mo_ldd.ldd_flags & (LDD_F_VIRGIN | LDD_F_WRITECONF)) {
+       if (mop->mo_ldd.ldd_flags & (LDD_F_VIRGIN | LDD_F_WRITECONF |
+           LDD_F_NO_LOCAL_LOGS)) {
                (void)osd_label_lustre(mop);
        } else {
                struct lustre_disk_data ldd;