Merge branch 'master' into lxz/test_tgt_nr

Required-githooks: true

ci/jira_query.py

@@ -27,8 +27,8 @@
 # Expected components from the commit message, and directory in src/, src/client or utils/ is also
 # valid.  We've never checked/enforced these before so there have been a lot of values used in the
 # past.
-VALID_COMPONENTS = ('agent', 'build', 'ci', 'csum', 'doc', 'gha', 'il', 'md', 'mercury', 'pil4dfs',
-                    'swim', 'test', 'tools')
+VALID_COMPONENTS = ('agent', 'build', 'ci', 'csum', 'doc', 'gha', 'il', 'md', 'mercury',
+                    'packaging', 'pil4dfs', 'swim', 'test', 'tools')
 
 # Expected ticket prefix.
 VALID_TICKET_PREFIX = ('DAOS', 'CORCI', 'SRE')

debian/changelog

@@ -1,3 +1,9 @@
+daos (2.5.100-7) unstable; urgency=medium
+  [ Brian J. Murrell ]
+  * NOOP change to keep in parity with RPM version
+
+ -- Brian J. Murrell <brian.murrell@intel.com>  Fri, 07 Jul 2023 16:05:01 -0400
+
 daos (2.5.100-6) unstable; urgency=medium
   [Michael MacDonald]
   * Add golang-go as a tests dependency for dfuse/daos_build.py

docs/admin/troubleshooting.md

@@ -463,6 +463,15 @@ fabric_iface_port: 31316
 # engine 1
 fabric_iface_port: 31416
 ```
+### daos_agent cache of engine URIs is stale
+
+The `daos_agent` cache may become invalid if `daos_engine` processes restart with different
+configurations or IP addresses, or if the DAOS system is reformatted.
+If this happens, the `daos` tool (as well as other I/O or `libdaos` operations) may return
+`-DER_BAD_TARGET` (-1035) errors.
+
+To resolve the issue, a privileged user may send a `SIGUSR2` signal to the `daos_agent` process to
+force an immediate cache refresh.
 
 ## Diagnostic and Recovery Tools
 

docs/user/filesystem.md

@@ -701,6 +701,30 @@ These are two command line options to control the DFuse process itself.
 
 These will affect all containers accessed via DFuse, regardless of any container attributes.
 
+### Managing memory usage and disconnecting from containers
+
+DFuse can be instructed to evict paths from local memory which drops any open handles on containers
+or pools as well as reducing the working set size and memory consumption.  This is an asynchronous
+operation and there is no automatic way to tell if it's completed.  In addition, any lookup of the
+path specified in the eviction call will cause a new lookup and prevent the eviction from
+completing.
+
+Paths can be requested for eviction from dfuse using the `daos filesystem evict` command.  This does
+not change any data that is stored in DAOS in any way but rather releases local resources.  This
+command will return the inode number of the path as well as key dfuse metrics.
+
+DFuse metrics can be queried with the `daos filesystem query` command which takes an optional
+`--inode` parameter.  This will return information on the number of inodes held in memory, the
+number of open files as well as the number of pools and containers that DFuse is connected to.  If
+the `--inode` option is given then this command will also report if the inode is in memory or not.
+
+Together these two commands can be used to request eviction of a path and to poll for its release,
+although lookups from other processes might block the eviction process.
+
+If `daos filesystem evict` is passed the root of the DFuse mount then the path itself cannot be
+evicted - in this case all top-level entries in the directory are evicted instead and no inode
+number is returned.
+
 ### Permissions
 
 DFuse can serve data from any user's container, but needs appropriate permissions in order to do

site_scons/env_modules.py

@@ -60,6 +60,7 @@ def _module_func(self, command, *arguments):  # pylint: disable=no-self-use
         except OSError as error:
             if error.errno == errno.ENOENT:
                 return None, None
+            raise
 
         stdout, stderr = proc.communicate()
 
@@ -189,6 +190,7 @@ def load_mpi(mpi):
             print("Error running update-alternatives")
             if error.errno == errno.ENOENT:
                 return False
+            raise
         for line in proc.stdout.readlines():
             if line.startswith(b"Value:"):
                 if line[line.rfind(b".") + 1:-1].decode() == mpi:

src/bio/bio_buffer.c

@@ -1859,16 +1859,19 @@ alloc_copy_desc(struct bio_io_context *ioctxt, struct umem_instance *umem,
 	return NULL;
 }
 
-struct bio_copy_desc *
+int
 bio_copy_prep(struct bio_io_context *ioctxt, struct umem_instance *umem,
-	      struct bio_sglist *bsgl_src, struct bio_sglist *bsgl_dst)
+	      struct bio_sglist *bsgl_src, struct bio_sglist *bsgl_dst,
+	      struct bio_copy_desc **desc)
 {
 	struct bio_copy_desc	*copy_desc;
-	int			 rc;
+	int			 rc, ret;
 
 	copy_desc = alloc_copy_desc(ioctxt, umem, bsgl_src, bsgl_dst);
-	if (copy_desc == NULL)
-		return NULL;
+	if (copy_desc == NULL) {
+		*desc = NULL;
+		return -DER_NOMEM;
+	}
 
 	rc = bio_iod_prep(copy_desc->bcd_iod_src, BIO_CHK_TYPE_LOCAL, NULL, 0);
 	if (rc)
@@ -1877,15 +1880,17 @@ bio_copy_prep(struct bio_io_context *ioctxt, struct umem_instance *umem,
 	copy_desc->bcd_iod_dst->bd_copy_dst = 1;
 	rc = bio_iod_prep(copy_desc->bcd_iod_dst, BIO_CHK_TYPE_LOCAL, NULL, 0);
 	if (rc) {
-		rc = bio_iod_post(copy_desc->bcd_iod_src, 0);
-		D_ASSERT(rc == 0);
+		ret = bio_iod_post(copy_desc->bcd_iod_src, 0);
+		D_ASSERT(ret == 0);
 		goto free;
 	}
 
-	return copy_desc;
+	*desc = copy_desc;
+	return 0;
 free:
 	free_copy_desc(copy_desc);
-	return NULL;
+	*desc = NULL;
+	return rc;
 }
 
 int
@@ -1950,9 +1955,9 @@ bio_copy(struct bio_io_context *ioctxt, struct umem_instance *umem,
 	struct bio_copy_desc	*copy_desc;
 	int			 rc;
 
-	copy_desc = bio_copy_prep(ioctxt, umem, bsgl_src, bsgl_dst);
-	if (copy_desc == NULL)
-		return -DER_NOMEM;
+	rc = bio_copy_prep(ioctxt, umem, bsgl_src, bsgl_dst, &copy_desc);
+	if (rc)
+		return rc;
 
 	rc = bio_copy_run(copy_desc, copy_size, csum_desc);
 	rc = bio_copy_post(copy_desc, rc);

src/bio/bio_context.c

@@ -193,7 +193,7 @@ blob_wait_completion(struct bio_xs_context *xs_ctxt, struct blob_cp_arg *ba)
 	} else {
 		rc = ABT_eventual_wait(ba->bca_eventual, NULL);
 		if (rc != ABT_SUCCESS)
-			D_ERROR("ABT eventual wait failed. %d", rc);
+			D_ERROR("ABT eventual wait failed. %d\n", rc);
 	}
 }
 
@@ -967,13 +967,13 @@ int bio_mc_open(struct bio_xs_context *xs_ctxt, uuid_t pool_id,
 close_wal_ioctxt:
 	rc1 = bio_ioctxt_close(bio_mc->mc_wal);
 	if (rc1)
-		D_ERROR("Failed to close wal ioctxt. %d", rc1);
+		D_ERROR("Failed to close wal ioctxt. %d\n", rc1);
 close_meta:
 	meta_close(bio_mc);
 close_meta_ioctxt:
 	rc1 = bio_ioctxt_close(bio_mc->mc_meta);
 	if (rc1)
-		D_ERROR("Failed to close meta ioctxt. %d", rc1);
+		D_ERROR("Failed to close meta ioctxt. %d\n", rc1);
 free_mem:
 	D_FREE(bio_mc);
 

src/bio/bio_device.c

@@ -748,7 +748,7 @@ led_device_action(void *ctx, struct spdk_pci_device *pci_device)
 		break;
 	case CTL__LED_ACTION__RESET:
 		/* Reset intercepted earlier in call-stack and converted to set */
-		D_ERROR("Reset action is not supported");
+		D_ERROR("Reset action is not supported\n");
 		opts->status = -DER_INVAL;
 		return;
 	default:

src/bio/bio_xstream.c

@@ -317,6 +317,13 @@ bio_nvme_init(const char *nvme_conf, int numa_node, unsigned int mem_size,
 		goto free_cond;
 	}
 
+	/*
+	 * Let's keep using large cluster size(1GB) for pmem mode, the SPDK blobstore
+	 * loading time is unexpected long for smaller cluster size(32MB), see DAOS-13694.
+	 */
+	if (!bio_nvme_configured(SMD_DEV_TYPE_META))
+		nvme_glb.bd_bs_opts.cluster_sz = (1UL << 30);	/* 1GB */
+
 	D_INFO("MD on SSD is %s\n",
 	       bio_nvme_configured(SMD_DEV_TYPE_META) ? "enabled" : "disabled");
 
@@ -954,7 +961,7 @@ init_bio_bdevs(struct bio_xs_context *ctxt)
 
 	D_ASSERT(!is_server_started());
 	if (spdk_bdev_first() == NULL) {
-		D_ERROR("No SPDK bdevs found!");
+		D_ERROR("No SPDK bdevs found!\n");
 		rc = -DER_NONEXIST;
 	}
 

src/cart/crt_corpc.c

@@ -94,6 +94,7 @@ crt_corpc_initiate(struct crt_rpc_priv *rpc_priv)
 	struct crt_grp_gdata	*grp_gdata;
 	struct crt_grp_priv	*grp_priv;
 	struct crt_corpc_hdr	*co_hdr;
+	int			 src_timeout;
 	bool			 grp_ref_taken = false;
 	int			 rc = 0;
 
@@ -121,6 +122,10 @@ crt_corpc_initiate(struct crt_rpc_priv *rpc_priv)
 		}
 	}
 
+	/* Inherit a timeout from a source */
+	src_timeout = rpc_priv->crp_req_hdr.cch_src_timeout;
+	rpc_priv->crp_timeout_sec = src_timeout;
+
 	rc = crt_corpc_info_init(rpc_priv, grp_priv, grp_ref_taken,
 				 co_hdr->coh_filter_ranks,
 				 co_hdr->coh_grp_ver /* grp_ver */,
@@ -874,6 +879,7 @@ crt_corpc_req_hdlr(struct crt_rpc_priv *rpc_priv)
 		child_rpc_priv = container_of(child_rpc, struct crt_rpc_priv,
 					      crp_pub);
 
+		child_rpc_priv->crp_timeout_sec = rpc_priv->crp_timeout_sec;
 		corpc_add_child_rpc(rpc_priv, child_rpc_priv);
 
 		child_rpc_priv->crp_grp_priv = co_info->co_grp_priv;

src/cart/crt_group.c

@@ -1826,24 +1826,23 @@ crt_group_config_path_set(const char *path)
 	int rc;
 
 	if (path == NULL) {
-		D_ERROR("path can't be NULL");
+		D_ERROR("path can't be NULL\n");
 		return -DER_INVAL;
 	}
 
 	if (strlen(path) >= CRT_MAX_ATTACH_PREFIX) {
-		D_ERROR("specified path must be fewer than %d characters",
-			CRT_MAX_ATTACH_PREFIX);
+		D_ERROR("specified path must be fewer than %d characters\n", CRT_MAX_ATTACH_PREFIX);
 		return -DER_INVAL;
 	}
 
 	rc = stat(path, &buf);
 	if (rc != 0) {
-		D_ERROR("bad path specified: %s", path);
+		D_ERROR("bad path specified: %s\n", path);
 		return d_errno2der(errno);
 	}
 
 	if (!S_ISDIR(buf.st_mode)) {
-		D_ERROR("not a directory: %s", path);
+		D_ERROR("not a directory: %s\n", path);
 		return -DER_NOTDIR;
 	}
 

src/cart/crt_hg_proc.c

@@ -542,6 +542,7 @@ crt_proc_in_common(crt_proc_t proc, crt_rpc_input_t *data)
 						);
 			hdr->cch_dst_tag = rpc_priv->crp_pub.cr_ep.ep_tag;
 
+			hdr->cch_src_timeout = rpc_priv->crp_timeout_sec;
 			if (crt_is_service()) {
 				hdr->cch_src_rank =
 					crt_grp_priv_get_primary_rank(

src/cart/crt_iv.c

@@ -1917,7 +1917,7 @@ crt_hdlr_iv_sync_aux(void *arg)
 		break;
 
 	default:
-		D_ERROR("Unknown event type %#x", sync_type->ivs_event);
+		D_ERROR("Unknown event type %#x\n", sync_type->ivs_event);
 		D_GOTO(exit, rc = -DER_INVAL);
 		break;
 	}
@@ -2551,8 +2551,7 @@ handle_ivupdate_response(const struct crt_cb_info *cb_info)
 
 		/* Respond back to child; might fail if child is not alive */
 		if (crt_reply_send(iv_info->uci_child_rpc) != DER_SUCCESS)
-			D_ERROR("Failed to respond on rpc: %p",
-				iv_info->uci_child_rpc);
+			D_ERROR("Failed to respond on rpc: %p\n", iv_info->uci_child_rpc);
 
 		/* ADDREF done in crt_hdlr_iv_update */
 		RPC_PUB_DECREF(iv_info->uci_child_rpc);
@@ -3306,8 +3305,11 @@ crt_iv_update_internal(crt_iv_namespace_t ivns, uint32_t class_id,
 		D_GOTO(exit, rc);
 	}
 
-	rc = iv_ops->ivo_on_get(ivns, iv_key,
-				0, CRT_IV_PERM_WRITE, NULL, &priv);
+	rc = iv_ops->ivo_on_get(ivns, iv_key, 0, CRT_IV_PERM_WRITE, NULL, &priv);
+	if (rc != 0) {
+		D_ERROR("ivo_on_get(): " DF_RC, DP_RC(rc));
+		D_GOTO(exit, rc);
+	}
 
 	if (iv_value != NULL)
 		rc = iv_ops->ivo_on_update(ivns, iv_key, 0,

src/cart/crt_rpc.c

@@ -1897,6 +1897,28 @@ crt_req_dst_tag_get(crt_rpc_t *rpc, uint32_t *tag)
 	return rc;
 }
 
+int
+crt_req_src_timeout_get(crt_rpc_t *rpc, uint16_t *timeout)
+{
+	struct crt_rpc_priv	*rpc_priv = NULL;
+	int			rc = 0;
+
+	if (rpc == NULL) {
+		D_ERROR("NULL rpc passed\n");
+		D_GOTO(out, rc = -DER_INVAL);
+	}
+
+	if (timeout == NULL) {
+		D_ERROR("NULL timeout passed\n");
+		D_GOTO(out, rc = -DER_INVAL);
+	}
+
+	rpc_priv = container_of(rpc, struct crt_rpc_priv, crp_pub);
+	*timeout = rpc_priv->crp_req_hdr.cch_src_timeout;
+out:
+	return rc;
+}
+
 int
 crt_register_hlc_error_cb(crt_hlc_error_cb event_handler, void *arg)
 {

src/cart/crt_rpc.h

@@ -65,12 +65,15 @@ struct crt_common_hdr {
 	d_rank_t	cch_dst_rank;
 	/* originator rank in default primary group */
 	d_rank_t	cch_src_rank;
-	/* tag to which rpc request was sent to */
-	uint32_t	cch_dst_tag;
+	/* destination tag */
+	uint16_t	cch_dst_tag;
+	/* source timeout, to be replaced by deadline eventually */
+	uint16_t	cch_src_timeout;
 	/* used in crp_reply_hdr to propagate rpc failure back to sender */
 	uint32_t	cch_rc;
 };
 
+
 typedef enum {
 	RPC_STATE_INITED = 0x36,
 	RPC_STATE_QUEUED, /* queued for flow controlling */

src/client/api/job.c

@@ -21,7 +21,7 @@ craft_default_jobid(char **jobid)
 
 	ret = uname(&name);
 	if (ret) {
-		D_ERROR("Uname to get uname for creating default jobid");
+		D_ERROR("Uname to get uname for creating default jobid\n");
 		return daos_errno2der(errno);
 	}
 

src/client/dfs/dfs.c

@@ -3156,7 +3156,7 @@ dfs_obj_set_oclass(dfs_t *dfs, dfs_obj_t *obj, int flags, daos_oclass_id_t cid)
 	rc = daos_obj_update(oh, DAOS_TX_NONE, DAOS_COND_DKEY_UPDATE, &dkey, 1,
 			     &iod, &sgl, NULL);
 	if (rc) {
-		D_ERROR("Failed to update object class ("DF_RC")\n", DP_RC(rc));
+		D_ERROR("Failed to update object class: " DF_RC "\n", DP_RC(rc));
 		D_GOTO(out, rc = daos_der2errno(rc));
 	}
 
@@ -3206,7 +3206,7 @@ set_chunk_size(dfs_t *dfs, dfs_obj_t *obj, daos_size_t csize)
 	rc = daos_obj_update(oh, DAOS_TX_NONE, DAOS_COND_DKEY_UPDATE, &dkey, 1,
 			     &iod, &sgl, NULL);
 	if (rc) {
-		D_ERROR("Failed to update chunk size ("DF_RC")\n", DP_RC(rc));
+		D_ERROR("Failed to update chunk size: " DF_RC "\n", DP_RC(rc));
 		D_GOTO(out, rc = daos_der2errno(rc));
 	}
 

src/client/dfs/duns.c

@@ -757,7 +757,7 @@ create_cont(daos_handle_t poh, struct duns_attr_t *attrp, bool create_with_label
 
 		prop = daos_prop_alloc(nr);
 		if (prop == NULL) {
-			D_ERROR("Failed to allocate container prop.");
+			D_ERROR("Failed to allocate container prop.\n");
 			return ENOMEM;
 		}
 		if (attrp->da_props != NULL) {

src/client/dfuse/dfuse.h

@@ -140,6 +140,8 @@ struct dfuse_obj_hdl {
 	bool                      doh_kreaddir_started;
 	/* Set to true if readdir calls reach EOF made on this handle */
 	bool                      doh_kreaddir_finished;
+
+	bool                      doh_evict_on_close;
 };
 
 /* Readdir support.
@@ -707,8 +709,7 @@ struct fuse_lowlevel_ops dfuse_ops;
 					strerror(-__rc));                                          \
 	} while (0)
 
-#define DFUSE_REPLY_IOCTL(desc, req, arg)			\
-	DFUSE_REPLY_IOCTL_SIZE(desc, req, &(arg), sizeof(arg))
+#define DFUSE_REPLY_IOCTL(desc, req, arg) DFUSE_REPLY_IOCTL_SIZE(desc, req, &(arg), sizeof(arg))
 
 /**
  * Inode handle.