Now you are talking about the possibility to save some memory on a 1-node system by e.g. introducing #ifdefs in the code to disable code paths that deal with redundancy. I can agree that in some very resource-constrained embedded system this could make sense, but on the other hand, this is not good from a testing perspective. You would have two separate builds that needs to be tested. From a developer's perspective, it is not good either. #ifdefs will mess up the code, and the developers also need to build and test the code twice. And what happens when you get the second feature, which is also enabled and disabled using #ifdefs. You will get four possible builds. With the third feature, eight possible builds.
On the other hand, a redundant system must work also on a single node. Otherwise it wouldn't be redundant - i.e. if you /require/ two nodes for the system to operate then you have a 2-node system which is less available than just one single node. If any of the two nodes fails then the whole system fails. So we must in any case test that we can run on a single node without problems. Unless we have a very strong use-case for saving those extra bytes of code I would not like to introduce a build-time option for disabling redundancy. We could introduce a run-time option for disallowing single-node configurations. In fact, I would argue in such a case that it would be an option to set the minimum allowed number of nodes in the system - i.e. the minimum size you can scale down to. It doesn't necessarily have to be two; maybe someone wishes to disallow any configuration with less than five nodes. regards, Anders Widell On 04/01/2016 03:05 PM, Mathivanan Naickan Palanivelu wrote: > We could tie the symmetry to the 'deploy-standlone' option chosen. > i.e. if 'deploy-standone' feature is enabled don't expect(wherever) standbys > including to not expect > things like setting up shared file system and all code flows that expect a > second/alternate to take over, etc! > I think this has to be thought through and not just considered as a matter of > 'allowing a configuration change' for > a tool to work. > It is another thing that we should discussed this when the asymmetry in > configuration got created! > If it's about reflecting the 'state' then there is sufficient states to > convey the status of that > application. > > >> -----Original Message----- >> From: Anders Widell [mailto:anders.wid...@ericsson.com] >> Sent: Friday, April 01, 2016 5:49 PM >> To: Mathivanan Naickan Palanivelu; Praveen Malviya; >> gary....@dektech.com.au; hans.nordeb...@ericsson.com; Nagendra Kumar >> Cc: opensaf-devel@lists.sourceforge.net >> Subject: Re: [devel] [PATCH 1 of 1] amf: Support AMF configurations >> containing more than two OpenSAF 2N SUs [#79] >> >> Today it is perfectly possible to configure a 1-node system without using any >> build-time option. The problem today is that if you expand a 1-node system to >> a 2-node system, then you cannot later shrink it back into a 1-node system >> again. This asymmetry that you cannot go back to where you came from >> makes little sense to me. Suppose that the second node has been physically >> removed, so that you really have only one single node. What harm would it do >> to update the OpenSAF configuration to reflect this fact of reality? >> >> regards, >> Anders Widell >> >> On 04/01/2016 12:54 PM, Mathivanan Naickan Palanivelu wrote: >>> Ofcourse there could be applications(in different problem domains) that >> could be configured to run in standalone mode or in HA mode. >>> Such applications could still be configured(in AMF) to be run on just 1 >> OpenSAF node, nothing in OpenSAF stops standalone applications today! >>> But, I don't think it is necessary to trickle down an application's >>> standalone >> configuration into OpenSAF's configuration too! >>> Somehow this topic has come up again, and I don't understand the >>> requirement behind why a user's tool expects HA middleware also to be >> configured in standalone mode! >>> If we still want to satisfy such(adhoc?) tools, then we could do that >>> by bringing in a 'standalone' deployment option in OpenSAF, i.e. >>> ./configure --deploy-standone. When this option is enable then immxml- >> tools, AMF and any other OpenSAF service would not expect a mandatory >> STANDBY to exist, thereby also allow deletion or any other operation of its >> interest! >>> Cheers, >>> Mathi. >>> >>>> -----Original Message----- >>>> From: praveen malviya >>>> Sent: Friday, April 01, 2016 3:52 PM >>>> To: Anders Widell; gary....@dektech.com.au; >>>> hans.nordeb...@ericsson.com; Nagendra Kumar >>>> Cc: opensaf-devel@lists.sourceforge.net >>>> Subject: Re: [devel] [PATCH 1 of 1] amf: Support AMF configurations >>>> containing more than two OpenSAF 2N SUs [#79] >>>> >>>> Please see my comments inline below: >>>> >>>> On 31-Mar-16 6:50 PM, Anders Widell wrote: >>>>> Yes I added the check in the admin op as you suggested. But I don't >>>>> fully agree that the same check should be done when removing system >>>>> controller nodes. With the introduction of this feature, we are >>>>> starting to move away from the concept of different node types >>>>> (controller / payload). Indeed, I removed the "type" member variable >>>>> from the node class in the latest patch. >>>> [Praveen] >>>> Yes, all nodes can be controller nodes or rather all nodes are the >>>> same except their roles. But that does not change the architecture of >>>> OpenSAF w.r.t the redundancy model. i.e. The configuration is still 2N >> redundancy model. >>>> i.e. The smallest opensaf sized cluster (without payloads) would >>>> still be configured either of the two options as below: >>>> (a) immxml-clustersize -s 2 -p0 >>>> During scale out, for spare addition atleast one standy has to exist >>>> before proceeding to configure the rest of the cluster nodes as >>>> standbys >>>> (OR) >>>> (b) immxml-clustersize -s 3 -p0 >>>> Obviously the 3rd node and all other nodes added after the two nodes >>>> would act as spares. >>>> >>>>> To preserve backwards compatibility, I can agree to have this check >>>>> on systems that are configured with both system controller nodes and >>>>> payload nodes (as in your example with immxml-clustersize). This >>>>> would mean that AMF will reject removal of any of the last two >>>>> system controller nodes - IF the cluster has payload nodes. What do >>>>> you think >>>> [Praveen] >>>> That is the normal case anyhow today. Only difference we have to make >>>> now is to allow deletion of spare nodes whether there are payloads or not. >>>> >>>> Thanks, >>>> Praveen. >>>> >>>>> about this approach? I am not sure how easy this would be to >>>>> implement, but I can give it a try. >>>>> >>>>> regards, >>>>> Anders Widell >>>>> >>>>> On 03/31/2016 03:05 PM, praveen malviya wrote: >>>>>> Hi, >>>>>> >>>>>> In the diff patch, I have seen that admin operation on MW 2N SU is >>>>>> not allowed when more than 2 SUs are configured which translates to >>>>>> the fact that system is running with spare controllers. >>>>>> >>>>>> A similar type of check is needed for deletion of controller >>>>>> configuration from AMF. The check would not allow deletion of >>>>>> controller if only two of them remained. It is inline with the >>>>>> current OpenSAF implementation with the fact that we do not allow >>>>>> any payload to get configured when user tries to generate imm.xml >>>>>> with single controller and a given no. of payload because such a >>>>>> configuration does not provide redundancy to payloads. >>>>>> >>>>>> Note: ./immxml-clustersize -s 1 -p 1 >>>>>> error: Two SC's is required for clusters with payloads. Exiting! >>>>>> >>>>>> >>>>>> Thanks, >>>>>> Praveen >>>>>> >>>>>> On 31-Mar-16 2:05 AM, Anders Widell wrote: >>>>>>> Here is a patch that addresses the review comments from Hans and >>>>>>> Praveen. It should be applied on top of the AMF patch that was >>>>>>> sent out for review. >>>>>>> >>>>>>> thanks, >>>>>>> Anders Widell >>>>>>> >>>>>>> On 03/30/2016 04:35 PM, Anders Widell wrote: >>>>>>>> Hi! >>>>>>>> >>>>>>>> See my replies inline, marked [AndersW]. >>>>>>>> >>>>>>>> regards, >>>>>>>> Anders Widell >>>>>>>> >>>>>>>> On 03/17/2016 11:32 AM, praveen malviya wrote: >>>>>>>>> Hi Anders, >>>>>>>>> >>>>>>>>> Please find some comments and queries inline with [Praveen] >>>>>>>>> >>>>>>>>> Thanks, >>>>>>>>> Praveen >>>>>>>>> >>>>>>>>> >>>>>>>>> On 29-Feb-16 8:44 PM, Anders Widell wrote: >>>>>>>>>> osaf/services/saf/amf/amfd/clm.cc | 21 +++++- >>>>>>>>>> osaf/services/saf/amf/amfd/include/amfd.h | 2 + >>>>>>>>>> osaf/services/saf/amf/amfd/include/cb.h | 1 + >>>>>>>>>> osaf/services/saf/amf/amfd/include/role.h | 2 + >>>>>>>>>> osaf/services/saf/amf/amfd/main.cc | 78 >>>>>>>>>> ++++++--------------------- >>>>>>>>>> osaf/services/saf/amf/amfd/ndfsm.cc | 9 ++- >>>>>>>>>> osaf/services/saf/amf/amfd/node.cc | 7 +-- >>>>>>>>>> osaf/services/saf/amf/amfd/role.cc | 86 >>>>>>>>>> ++++++++++++++++++++++++++++++- >>>>>>>>>> osaf/services/saf/amf/amfd/sgproc.cc | 11 +++- >>>>>>>>>> osaf/services/saf/amf/amfnd/clm.cc | 27 ++++++-- >>>>>>>>>> 10 files changed, 160 insertions(+), 84 deletions(-) >>>>>>>>>> >>>>>>>>>> >>>>>>>>>> Add support for configuring the system with more than two >>>>>>>>>> OpenSAF >>>> 2N >>>>>>>>>> SUs. In >>>>>>>>>> particular, this means that all OpenSAF directors must support >>>>>>>>>> starting up and running without (initially) getting any >>>>>>>>>> assignment from AMF. >>>>>>>>>> Locking of >>>>>>>>>> an OpenSAF 2N SU is currently not supported on a system >>>>>>>>>> configured with more than two OpenSAF 2N SUs. >>>>>>>>> [Praveen] This patch does not contain any change for any >>>>>>>>> restricton on locking of OpenSAF 2N SU as mentioned above. >>>>>>>> [AndersW] No. This restriction will be documented. Do you think >>>>>>>> we need to add checks for this case in the admin op as well? >>>>>>>>>> diff --git a/osaf/services/saf/amf/amfd/clm.cc >>>>>>>>>> b/osaf/services/saf/amf/amfd/clm.cc >>>>>>>>>> --- a/osaf/services/saf/amf/amfd/clm.cc >>>>>>>>>> +++ b/osaf/services/saf/amf/amfd/clm.cc >>>>>>>>>> @@ -21,8 +21,7 @@ >>>>>>>>>> #include <amfd.h> >>>>>>>>>> #include <clm.h> >>>>>>>>>> #include <node.h> >>>>>>>>>> - >>>>>>>>>> -static SaVersionT clmVersion = { 'B', 4, 1 }; >>>>>>>>>> +#include "osaf_time.h" >>>>>>>>>> >>>>>>>>>> static void clm_node_join_complete(AVD_AVND *node) >>>>>>>>>> { >>>>>>>>>> @@ -392,9 +391,21 @@ SaAisErrorT avd_clm_init(void) >>>>>>>>>> SaAisErrorT error = SA_AIS_OK; >>>>>>>>>> >>>>>>>>>> TRACE_ENTER(); >>>>>>>>>> - error = saClmInitialize_4(&avd_cb->clmHandle, &clm_callbacks, >>>>>>>>>> &clmVersion); >>>>>>>>>> - if (SA_AIS_OK != error) { >>>>>>>>>> - LOG_ER("Failed to initialize with CLM %u", error); >>>>>>>>>> + for (;;) { >>>>>>>>>> + SaVersionT Version = { 'B', 4, 1 }; >>>>>>>>>> + error = saClmInitialize_4(&avd_cb->clmHandle, >>>>>>>>>> &clm_callbacks, &Version); >>>>>>>>>> + if (error == SA_AIS_ERR_TRY_AGAIN || >>>>>>>>>> + error == SA_AIS_ERR_TIMEOUT || >>>>>>>>>> + error == SA_AIS_ERR_UNAVAILABLE) { >>>>>>>>>> + if (error != SA_AIS_ERR_TRY_AGAIN) { >>>>>>>>>> + LOG_WA("saClmInitialize_4 returned %u", >>>>>>>>>> + (unsigned) error); >>>>>>>>>> + } >>>>>>>>>> + osaf_nanosleep(&kHundredMilliseconds); >>>>>>>>>> + continue; >>>>>>>>>> + } >>>>>>>>>> + if (error == SA_AIS_OK) break; >>>>>>>>>> + LOG_ER("Failed to Initialize with CLM: %u", error); >>>>>>>>>> goto done; >>>>>>>>>> } >>>>>>>>>> error = saClmSelectionObjectGet(avd_cb->clmHandle, >>>>>>>>>> &avd_cb->clm_sel_obj); >>>>>>>>>> diff --git a/osaf/services/saf/amf/amfd/include/amfd.h >>>>>>>>>> b/osaf/services/saf/amf/amfd/include/amfd.h >>>>>>>>>> --- a/osaf/services/saf/amf/amfd/include/amfd.h >>>>>>>>>> +++ b/osaf/services/saf/amf/amfd/include/amfd.h >>>>>>>>>> @@ -33,6 +33,7 @@ >>>>>>>>>> #ifndef AVD_H >>>>>>>>>> #define AVD_H >>>>>>>>>> >>>>>>>>>> +#include <stdint.h> >>>>>>>>>> #include "logtrace.h" >>>>>>>>>> >>>>>>>>>> #include "amf.h" >>>>>>>>>> @@ -65,5 +66,6 @@ >>>>>>>>>> #include "ckpt_msg.h" >>>>>>>>>> #include "ckpt_edu.h" >>>>>>>>>> #include "ckpt_updt.h" >>>>>>>>>> +#include "saAmf.h" >>>>>>>>>> >>>>>>>>>> #endif >>>>>>>>>> diff --git a/osaf/services/saf/amf/amfd/include/cb.h >>>>>>>>>> b/osaf/services/saf/amf/amfd/include/cb.h >>>>>>>>>> --- a/osaf/services/saf/amf/amfd/include/cb.h >>>>>>>>>> +++ b/osaf/services/saf/amf/amfd/include/cb.h >>>>>>>>>> @@ -207,6 +207,7 @@ typedef struct cl_cb_tag { >>>>>>>>>> SaClmHandleT clmHandle; >>>>>>>>>> SaSelectionObjectT clm_sel_obj; >>>>>>>>>> >>>>>>>>>> + bool fully_initialized; >>>>>>>>>> bool swap_switch; /* true - In middle of role switch. */ >>>>>>>>>> >>>>>>>>>> /** true when active services (IMM, LOG, NTF, etc.) >>>>>>>>>> exist diff --git a/osaf/services/saf/amf/amfd/include/role.h >>>>>>>>>> b/osaf/services/saf/amf/amfd/include/role.h >>>>>>>>>> --- a/osaf/services/saf/amf/amfd/include/role.h >>>>>>>>>> +++ b/osaf/services/saf/amf/amfd/include/role.h >>>>>>>>>> @@ -34,6 +34,8 @@ extern uint32_t amfd_switch_qsd_stdby(AV >>>>>>>>>> extern uint32_t amfd_switch_stdby_actv(AVD_CL_CB *cb); >>>>>>>>>> extern uint32_t amfd_switch_qsd_actv(AVD_CL_CB *cb); >>>>>>>>>> extern uint32_t amfd_switch_actv_qsd(AVD_CL_CB *cb); >>>>>>>>>> +extern uint32_t initialize_for_assignment(cl_cb_tag* cb, >>>>>>>>>> + SaAmfHAStateT >>>>>>>>>> +ha_state); >>>>>>>>>> >>>>>>>>>> #endif /* ROLE_H */ >>>>>>>>>> >>>>>>>>>> diff --git a/osaf/services/saf/amf/amfd/main.cc >>>>>>>>>> b/osaf/services/saf/amf/amfd/main.cc >>>>>>>>>> --- a/osaf/services/saf/amf/amfd/main.cc >>>>>>>>>> +++ b/osaf/services/saf/amf/amfd/main.cc >>>>>>>>>> @@ -56,6 +56,7 @@ >>>>>>>>>> #include <sutcomptype.h> >>>>>>>>>> #include <sutype.h> >>>>>>>>>> #include <su.h> >>>>>>>>>> +#include "osaf_utility.h" >>>>>>>>>> >>>>>>>>>> static const char* internal_version_id_ __attribute__ >>>>>>>>>> ((used)) = >>>>>>>>>> "@(#) $Id: " INTERNAL_VERSION_ID " $"; >>>>>>>>>> >>>>>>>>>> @@ -445,7 +446,8 @@ static void rda_cb(uint32_t notused, PCS >>>>>>>>>> >>>>>>>>>> if (((avd_cb->avail_state_avd == SA_AMF_HA_STANDBY) || >>>>>>>>>> (avd_cb->avail_state_avd == SA_AMF_HA_QUIESCED)) && >>>>>>>>>> - (cb_info->info.io_role == PCS_RDA_ACTIVE)) { >>>>>>>>>> + (cb_info->info.io_role == PCS_RDA_ACTIVE || >>>>>>>>>> + cb_info->info.io_role == PCS_RDA_STANDBY)) { >>>>>>>>>> >>>>>>>>>> uint32_t rc; >>>>>>>>>> AVD_EVT *evt; >>>>>>>>>> @@ -474,7 +476,6 @@ static uint32_t initialize(void) >>>>>>>>>> { >>>>>>>>>> AVD_CL_CB *cb = avd_cb; >>>>>>>>>> int rc = NCSCC_RC_FAILURE; >>>>>>>>>> - SaVersionT ntfVersion = { 'A', 0x01, 0x01 }; >>>>>>>>>> SaAmfHAStateT role; >>>>>>>>>> char *val; >>>>>>>>>> >>>>>>>>>> @@ -524,8 +525,13 @@ static uint32_t initialize(void) >>>>>>>>>> } >>>>>>>>>> >>>>>>>>>> cb->init_state = AVD_INIT_BGN; >>>>>>>>>> + cb->mbcsv_sel_obj = -1; >>>>>>>>>> + cb->imm_sel_obj = -1; >>>>>>>>>> + cb->clm_sel_obj = -1; >>>>>>>>>> + cb->fully_initialized = false; >>>>>>>>>> cb->swap_switch = false; >>>>>>>>>> cb->active_services_exist = true; >>>>>>>>>> + cb->mbcsv_sel_obj = -1; >>>>>>>>> [Praveen] Duplicate initialization, already done above. >>>>>>>> [Anders W] Will remove. >>>>>>>>>> cb->stby_sync_state = AVD_STBY_IN_SYNC; >>>>>>>>>> cb->sync_required = true; >>>>>>>>>> >>>>>>>>>> @@ -544,67 +550,20 @@ static uint32_t initialize(void) >>>>>>>>>> /* get the node id of the node on which the AVD is running. >>>>>>>>>> */ >>>>>>>>>> cb->node_id_avd = m_NCS_GET_NODE_ID; >>>>>>>>>> >>>>>>>>>> - if (avd_mds_init(cb) != NCSCC_RC_SUCCESS) { >>>>>>>>>> - LOG_ER("avd_mds_init FAILED"); >>>>>>>>>> - goto done; >>>>>>>>>> - } >>>>>>>>>> - >>>>>>>>>> - if (NCSCC_RC_FAILURE == avsv_mbcsv_register(cb)) { >>>>>>>>>> - LOG_ER("avsv_mbcsv_register FAILED"); >>>>>>>>>> - goto done; >>>>>>>>>> - } >>>>>>>>>> - >>>>>>>>>> - if (avd_clm_init() != SA_AIS_OK) { >>>>>>>>>> - LOG_EM("avd_clm_init FAILED"); >>>>>>>>>> - goto done; >>>>>>>>>> - } >>>>>>>>>> - >>>>>>>>>> - if (avd_imm_init(cb) != SA_AIS_OK) { >>>>>>>>>> - LOG_ER("avd_imm_init FAILED"); >>>>>>>>>> - goto done; >>>>>>>>>> - } >>>>>>>>>> - >>>>>>>>>> - if ((rc = saNtfInitialize(&cb->ntfHandle, nullptr, >>>>>>>>>> &ntfVersion)) != SA_AIS_OK) { >>>>>>>>>> - LOG_ER("saNtfInitialize Failed (%u)", rc); >>>>>>>>>> - rc = NCSCC_RC_FAILURE; >>>>>>>>>> - goto done; >>>>>>>>>> - } >>>>>>>>>> - >>>>>>>>>> if ((rc = rda_get_role(&role)) != NCSCC_RC_SUCCESS) { >>>>>>>>>> LOG_ER("rda_get_role FAILED"); >>>>>>>>>> goto done; >>>>>>>>>> } >>>>>>>>>> >>>>>>>>>> - cb->avail_state_avd = role; >>>>>>>>>> - >>>>>>>>>> - if (NCSCC_RC_SUCCESS != avd_mds_set_vdest_role(cb, role)) { >>>>>>>>>> - LOG_ER("avd_mds_set_vdest_role FAILED"); >>>>>>>>>> - goto done; >>>>>>>>>> - } >>>>>>>>>> - >>>>>>>>>> - if (NCSCC_RC_SUCCESS != avsv_set_ckpt_role(cb, role)) { >>>>>>>>>> - LOG_ER("avsv_set_ckpt_role FAILED"); >>>>>>>>>> - goto done; >>>>>>>>>> - } >>>>>>>>>> - >>>>>>>>>> if ((rc = rda_register_callback(0, rda_cb)) != >>>>>>>>>> NCSCC_RC_SUCCESS) { >>>>>>>>>> LOG_ER("rda_register_callback FAILED %u", rc); >>>>>>>>>> goto done; >>>>>>>>>> } >>>>>>>>>> >>>>>>>>>> - if (role == SA_AMF_HA_ACTIVE) { >>>>>>>>>> - rc = avd_active_role_initialization(cb, role); >>>>>>>>>> - if (rc != NCSCC_RC_SUCCESS) { >>>>>>>>>> - LOG_ER("avd_active_role_initialization FAILED"); >>>>>>>>>> - goto done; >>>>>>>>>> - } >>>>>>>>>> - } >>>>>>>>>> - else { >>>>>>>>>> - rc = avd_standby_role_initialization(cb); >>>>>>>>>> - if (rc != NCSCC_RC_SUCCESS) { >>>>>>>>>> - LOG_ER("avd_standby_role_initialization FAILED"); >>>>>>>>>> - goto done; >>>>>>>>>> - } >>>>>>>>>> + if ((rc = initialize_for_assignment(cb, role)) >>>>>>>>>> + != NCSCC_RC_SUCCESS) { >>>>>>>>>> + LOG_ER("initialize_for_assignment FAILED %u", >>>>>>>>>> + (unsigned) >>>>>>>>>> rc); >>>>>>>>>> + goto done; >>>>>>>>>> } >>>>>>>>>> >>>>>>>>>> rc = NCSCC_RC_SUCCESS; >>>>>>>>>> @@ -647,14 +606,13 @@ static void main_loop(void) >>>>>>>>>> fds[FD_TERM].events = POLLIN; >>>>>>>>>> fds[FD_MBX].fd = mbx_fd.rmv_obj; >>>>>>>>>> fds[FD_MBX].events = POLLIN; >>>>>>>>>> - fds[FD_MBCSV].fd = cb->mbcsv_sel_obj; >>>>>>>>>> - fds[FD_MBCSV].events = POLLIN; >>>>>>>>>> - fds[FD_CLM].fd = cb->clm_sel_obj; >>>>>>>>>> - fds[FD_CLM].events = POLLIN; >>>>>>>>>> - fds[FD_IMM].fd = cb->imm_sel_obj; // IMM fd must be last in >>>>>>>>>> array >>>>>>>>>> - fds[FD_IMM].events = POLLIN; >>>>>>>>>> - >>>>>>>>>> while (1) { >>>>>>>>>> + fds[FD_MBCSV].fd = cb->mbcsv_sel_obj; >>>>>>>>>> + fds[FD_MBCSV].events = POLLIN; >>>>>>>>>> + fds[FD_CLM].fd = cb->clm_sel_obj; >>>>>>>>>> + fds[FD_CLM].events = POLLIN; >>>>>>>>>> + fds[FD_IMM].fd = cb->imm_sel_obj; // IMM fd must be >>>>>>>>>> + last in >>>>>>>>>> array >>>>>>>>>> + fds[FD_IMM].events = POLLIN; >>>>>>>>>> >>>>>>>>>> if (cb->immOiHandle != 0) { >>>>>>>>>> fds[FD_IMM].fd = cb->imm_sel_obj; diff --git >>>>>>>>>> a/osaf/services/saf/amf/amfd/ndfsm.cc >>>>>>>>>> b/osaf/services/saf/amf/amfd/ndfsm.cc >>>>>>>>>> --- a/osaf/services/saf/amf/amfd/ndfsm.cc >>>>>>>>>> +++ b/osaf/services/saf/amf/amfd/ndfsm.cc >>>>>>>>>> @@ -190,8 +190,9 @@ void avd_nd_ncs_su_assigned(AVD_CL_CB >> *c >>>>>>>>>> TRACE_ENTER(); >>>>>>>>>> >>>>>>>>>> for (const auto& ncs_su : avnd->list_of_ncs_su) { >>>>>>>>>> - if ((ncs_su->list_of_susi == AVD_SU_SI_REL_NULL) || >>>>>>>>>> - (ncs_su->list_of_susi->fsm != AVD_SU_SI_STATE_ASGND)) >> { >>>>>>>>>> + if ((ncs_su->sg_of_su->curr_assigned_sus() < 2) && >>>>>>>>>> + ((ncs_su->list_of_susi == AVD_SU_SI_REL_NULL) || >>>>>>>>>> + (ncs_su->list_of_susi->fsm != >>>>>>>>>> + AVD_SU_SI_STATE_ASGND))) { >>>>>>>>>> TRACE_LEAVE(); >>>>>>>>>> /* this is an unassigned SU so no need to scan >>>>>>>>>> further return here. */ >>>>>>>>>> return; >>>>>>>>>> @@ -328,6 +329,10 @@ void >> avd_mds_avnd_down_evh(AVD_CL_CB >>>> *cb >>>>>>>>>> if (avd_cb->avail_state_avd == SA_AMF_HA_ACTIVE) { >>>>>>>>>> avd_node_failover(node); >>>>>>>>>> + // Update standby out of sync if standby sc goes down >>>>>>>>>> + if (eravd_cb->node_id_avd_oth == >>>>>>>>>> node->node_info.nodeId) { >>>>>>>>>> + cb->stby_sync_state = AVD_STBY_OUT_OF_SYNC; >>>>>>>>>> + } >>>>>>>>> [Praveen] Deep down in a function call in avd_node_failover(), >>>>>>>>> AMF marks standby status IN_SYNC in some special case. AMF >> marks >>>>>>>>> out of sync status whenever it gets sync request in chkop.cc so >>>>>>>>> it is not needed here. >>>>>>>> [AndersW] I don't remember why this was added. Will check and >>>> remove >>>>>>>> if it doesn't cause any problems. Isn't it a bit strange though, >>>>>>>> so say that the standby is in sync when in fact it is down? >>>>>>>>>> } else { >>>>>>>>>> /* Remove dynamic info for node but keep in >>>>>>>>>> nodeid tree. >>>>>>>>>> * Possibly used at the end of controller >>>>>>>>>> failover to diff --git a/osaf/services/saf/amf/amfd/node.cc >>>>>>>>>> b/osaf/services/saf/amf/amfd/node.cc >>>>>>>>>> --- a/osaf/services/saf/amf/amfd/node.cc >>>>>>>>>> +++ b/osaf/services/saf/amf/amfd/node.cc >>>>>>>>>> @@ -120,7 +120,7 @@ void AVD_AVND::initialize() { >>>>>>>>>> pg_csi_list = {}; >>>>>>>>>> pg_csi_list.order = NCS_DBLIST_ANY_ORDER; >>>>>>>>>> pg_csi_list.cmp_cookie = avsv_dblist_uns32_cmp; >>>>>>>>>> - type = AVSV_AVND_CARD_PAYLOAD; >>>>>>>>>> + type = AVSV_AVND_CARD_SYS_CON; >>>>>>>>> [Praveen] Initially keeping as PAYLOAD, AMFD later changes node >>>>>>>>> type of controller AMFNDs to SYS_CON after evaluating some >> parameters. >>>>>>>>> Above default is set to SYS_CON, there is no change in this >>>>>>>>> patch which sets payload as payload. In this way node type of a >>>>>>>>> payload will also become SYS_CON. >>>>>>>> [AndersW] I think actually the "type" member variable ought to be >>>>>>>> removed, since it is not needed. I can update the patch to remove >>>>>>>> this variable completely. >>>>>>>>>> rcv_msg_id = {}; >>>>>>>>>> snd_msg_id = {}; >>>>>>>>>> cluster_list_node_next = {}; @@ -486,11 +486,6 @@ static >>>>>>>>>> SaAisErrorT node_ccb_completed_de >>>>>>>>>> return SA_AIS_ERR_BAD_OPERATION; >>>>>>>>>> } >>>>>>>>>> >>>>>>>>>> - if (node->type == AVSV_AVND_CARD_SYS_CON) { >>>>>>>>>> - report_ccb_validation_error(opdata, "Cannot remove >>>>>>>>>> controller node"); >>>>>>>>>> - return SA_AIS_ERR_BAD_OPERATION; >>>>>>>>>> - } >>>>>>>>>> - >>>>>>>>> [Praveen] Based on some earliar discussion, I remember deletion >>>>>>>>> of sys controller is restricted. So check should modified to >>>>>>>>> reject the operation if there are only two system controlers in the >> system. >>>>>>>> [AndersW] Now you can create a cluster with many system >>>>>>>> controller nodes (all nodes could be controllers), so we can't >>>>>>>> have this restriction any longer. I don't see any reason to treat >>>>>>>> the two-node case in a special way. You can create a cluster >>>>>>>> consisting of only one single node, and then scale it out to a >>>>>>>> two-node cluster. Why shouldn't it be possible to scale it back to one >> single node again? >>>>>>>>>> /* Check to see that the node is in admin locked state >>>>>>>>>> before delete */ >>>>>>>>>> if (node->saAmfNodeAdminState != >>>>>>>>>> SA_AMF_ADMIN_LOCKED_INSTANTIATION) { >>>>>>>>>> report_ccb_validation_error(opdata, "Node '%s' is >>>>>>>>>> not locked instantiation", opdata->objectName.value); diff >>>>>>>>>> --git a/osaf/services/saf/amf/amfd/role.cc >>>>>>>>>> b/osaf/services/saf/amf/amfd/role.cc >>>>>>>>>> --- a/osaf/services/saf/amf/amfd/role.cc >>>>>>>>>> +++ b/osaf/services/saf/amf/amfd/role.cc >>>>>>>>>> @@ -46,6 +46,7 @@ >>>>>>>>>> #include <si_dep.h> >>>>>>>>>> #include "osaf_utility.h" >>>>>>>>>> #include "role.h" >>>>>>>>>> +#include "nid_api.h" >>>>>>>>>> >>>>>>>>>> extern pthread_mutex_t imm_reinit_mutex; >>>>>>>>>> >>>>>>>>>> @@ -73,7 +74,15 @@ void avd_role_change_evh(AVD_CL_CB *cb, >>>>>>>>>> AVD_ROLE_CHG_CAUSE_T cause = >>>>>>>>>> msg->msg_info.d2d_chg_role_req.cause; >>>>>>>>>> SaAmfHAStateT role = >>>>>>>>>> msg->msg_info.d2d_chg_role_req.role; >>>>>>>>>> >>>>>>>>>> - TRACE_ENTER2("cause=%u, role=%u", cause, role); >>>>>>>>>> + TRACE_ENTER2("cause=%u, role=%u, current_role=%u", cause, >>>> role, >>>>>>>>>> + cb->avail_state_avd); >>>>>>>>>> + >>>>>>>>>> + if ((status = initialize_for_assignment(cb, role)) >>>>>>>>>> + != NCSCC_RC_SUCCESS) { >>>>>>>>>> + LOG_ER("initialize_for_assignment FAILED %u", >>>>>>>>>> + (unsigned) status); >>>>>>>>>> + _exit(EXIT_FAILURE); >>>>>>>>>> + } >>>>>>>>>> >>>>>>>>>> if (cb->avail_state_avd == role) { >>>>>>>>>> goto done; >>>>>>>>>> @@ -128,6 +137,13 @@ void avd_role_change_evh(AVD_CL_CB >> *cb, >>>>>>>>>> } >>>>>>>>>> >>>>>>>>>> if ((cause == AVD_FAIL_OVER) && >>>>>>>>>> + (cb->avail_state_avd == SA_AMF_HA_QUIESCED) && (role >>>>>>>>>> + == >>>>>>>>>> SA_AMF_HA_STANDBY)) { >>>>>>>>>> + /* Fail-over Quiesced to Active */ >>>>>>>>>> + status = NCSCC_RC_SUCCESS; >>>>>>>>>> + goto done; >>>>>>>>>> + } >>>>>>>>> [Praveen] A quiesced controller never goes from quiesced to >>>>>>>>> stanby only in swichover and not in failover. >>>>>>>>> So comment must be /*Fail-over Quiesced to standby (spare >>>>>>>>> controller role change)*/ >>>>>>>> [AndersW] Will update the comment. >>>>>>>>>> + >>>>>>>>>> + if ((cause == AVD_FAIL_OVER) && >>>>>>>>>> (cb->avail_state_avd == SA_AMF_HA_QUIESCED) && (role >>>>>>>>>> == >>>>>>>>>> SA_AMF_HA_ACTIVE)) { >>>>>>>>>> /* Fail-over Quiesced to Active */ >>>>>>>>>> status = avd_role_failover_qsd_actv(cb, role); @@ >>>>>>>>>> -155,7 +171,73 @@ void avd_role_change_evh(AVD_CL_CB *cb, >>>>>>>>>> return; >>>>>>>>>> } >>>>>>>>>> >>>>>>>>>> - >> /********************************************************** >>>> ******************\ >>>>>>>>>> +uint32_t initialize_for_assignment(cl_cb_tag* cb, >>>>>>>>>> +SaAmfHAStateT >>>>>>>>>> ha_state) >>>>>>>>>> +{ >>>>>>>>>> + TRACE_ENTER2("ha_state = %d", static_cast<int>(ha_state)); >>>>>>>>>> + SaVersionT ntfVersion = {'A', 0x01, 0x01}; >>>>>>>>>> + uint32_t rc = NCSCC_RC_SUCCESS; >>>>>>>>>> + SaAisErrorT error; >>>>>>>>>> + if (cb->fully_initialized) goto done; >>>>>>>>>> + cb->avail_state_avd = ha_state; >>>>>>>>>> + if (ha_state == SA_AMF_HA_QUIESCED) { >>>>>>>>>> + if ((rc = nid_notify(const_cast<char*>("AMFD"), >>>>>>>>>> + NCSCC_RC_SUCCESS, nullptr)) != >>>>>>>>>> NCSCC_RC_SUCCESS) { >>>>>>>>>> + LOG_ER("nid_notify failed"); >>>>>>>>>> + } >>>>>>>>>> + goto done; >>>>>>>>>> + } >>>>>>>>>> + if ((rc = avd_mds_init(cb)) != NCSCC_RC_SUCCESS) { >>>>>>>>>> + LOG_ER("avd_mds_init FAILED"); >>>>>>>>>> + goto done; >>>>>>>>>> + } >>>>>>>>>> + if ((rc = avsv_mbcsv_register(cb)) != NCSCC_RC_SUCCESS) { >>>>>>>>>> + LOG_ER("avsv_mbcsv_register FAILED"); >>>>>>>>>> + goto done; >>>>>>>>>> + } >>>>>>>>>> + if (avd_clm_init() != SA_AIS_OK) { >>>>>>>>>> + LOG_EM("avd_clm_init FAILED"); >>>>>>>>>> + rc = NCSCC_RC_FAILURE; >>>>>>>>>> + goto done; >>>>>>>>>> + } >>>>>>>>>> + if (avd_imm_init(cb) != SA_AIS_OK) { >>>>>>>>>> + LOG_ER("avd_imm_init FAILED"); >>>>>>>>>> + rc = NCSCC_RC_FAILURE; >>>>>>>>>> + goto done; >>>>>>>>>> + } >>>>>>>>>> + if ((error = saNtfInitialize(&cb->ntfHandle, nullptr, >>>>>>>>>> &ntfVersion)) != >>>>>>>>>> + SA_AIS_OK) { >>>>>>>>>> + LOG_ER("saNtfInitialize Failed (%u)", error); >>>>>>>>>> + rc = NCSCC_RC_FAILURE; >>>>>>>>>> + goto done; >>>>>>>>>> + } >>>>>>>>>> + if ((rc = avd_mds_set_vdest_role(cb, ha_state)) != >>>>>>>>>> NCSCC_RC_SUCCESS) { >>>>>>>>>> + LOG_ER("avd_mds_set_vdest_role FAILED"); >>>>>>>>>> + goto done; >>>>>>>>>> + } >>>>>>>>>> + if ((rc = avsv_set_ckpt_role(cb, ha_state)) != >>>>>>>>>> NCSCC_RC_SUCCESS) { >>>>>>>>>> + LOG_ER("avsv_set_ckpt_role FAILED"); >>>>>>>>>> + goto done; >>>>>>>>>> + } >>>>>>>>>> + if (ha_state == SA_AMF_HA_ACTIVE) { >>>>>>>>>> + rc = avd_active_role_initialization(cb, ha_state); >>>>>>>>>> + if (rc != NCSCC_RC_SUCCESS) { >>>>>>>>>> + LOG_ER("avd_active_role_initialization FAILED"); >>>>>>>>>> + goto done; >>>>>>>>>> + } >>>>>>>>>> + } else if (ha_state == SA_AMF_HA_STANDBY) { >>>>>>>>>> + rc = avd_standby_role_initialization(cb); >>>>>>>>>> + if (rc != NCSCC_RC_SUCCESS) { >>>>>>>>>> + LOG_ER("avd_standby_role_initialization FAILED"); >>>>>>>>>> + goto done; >>>>>>>>>> + } >>>>>>>>>> + } >>>>>>>>>> + cb->fully_initialized = true; >>>>>>>>>> +done: >>>>>>>>>> + TRACE_LEAVE2("rc = %u", rc); >>>>>>>>>> + return rc; >>>>>>>>>> +} >>>>>>>>>> + >>>>>>>>>> >> +/********************************************************* >>>> ******************* >>>>>>>>>> \ >>>>>>>>>> * Function: avd_init_role_set >>>>>>>>>> * >>>>>>>>>> * Purpose: AVSV function to handle AVD's initial role setting. >>>>>>>>>> diff --git a/osaf/services/saf/amf/amfd/sgproc.cc >>>>>>>>>> b/osaf/services/saf/amf/amfd/sgproc.cc >>>>>>>>>> --- a/osaf/services/saf/amf/amfd/sgproc.cc >>>>>>>>>> +++ b/osaf/services/saf/amf/amfd/sgproc.cc >>>>>>>>>> @@ -1390,7 +1390,16 @@ void avd_su_si_assign_evh(AVD_CL_CB >>>> *cb, >>>>>>>>>> /* Since a NCS SU has been assigned >>>>>>>>>> trigger the node FSM. */ >>>>>>>>>> /* For (ncs_spec == SA_TRUE), su will >>>>>>>>>> not be external, so su >>>>>>>>>> will have node attached. */ >>>>>>>>>> - avd_nd_ncs_su_assigned(cb, >>>>>>>>>> susi->su->su_on_node); >>>>>>>>>> + for (AmfDb<uint32_t, >>>>>>>>>> + AVD_AVND>::const_iterator >>>>>>>>>> it = node_id_db->begin(); >>>>>>>>>> + it != node_id_db->end(); it++) { >>>>>>>>>> + AVD_AVND *node = >>>>>>>>>> const_cast<AVD_AVND*>((*it).second); >>>>>>>>>> + >>>>>>>>>> + if (node->node_state == >>>>>>>>>> AVD_AVND_STATE_NCS_INIT && node->adest != 0) { >>>>>>>>>> + avd_nd_ncs_su_assigned(cb, node); >>>>>>>>>> + } else { >>>>>>>>>> + TRACE("Node_state: %u adest: %" >>>>>>>>>> + PRIx64 >>>>>>>>>> " node not ready for assignments", node->node_state, node- >>>>> adest); >>>>>>>>>> + } >>>>>>>>>> + } >>>>>>>>> [Praveen] Could not understand this change? Since spare >>>>>>>>> controllers are also payloads, in which case adest can be 0. Is >>>>>>>>> this for headless case, there comp and su assignment information >>>>>>>>> comes before node >>>> up >>>>>>>>> message. >>>>>>>> [AndersW] This loop is needed to ensure set_leds is performed. It >>>>>>>> is sufficient that we have an active and a standby assignment for >>>>>>>> the OpenSAF 2N SU, so once we have that we need to loop over all >>>>>>>> nodes >>>> and >>>>>>>> perform set_leds if possible. >>>>>>>> >>>>>>>> I think the check that adest is non-zero was added by Hans, to >>>>>>>> fix some problem that might be related to headless if I remember >>>>>>>> correctly. @Hans, do you remember why this check was needed? >>>>>>>>>> } >>>>>>>>>> } >>>>>>>>>> } else { >>>>>>>>>> diff --git a/osaf/services/saf/amf/amfnd/clm.cc >>>>>>>>>> b/osaf/services/saf/amf/amfnd/clm.cc >>>>>>>>>> --- a/osaf/services/saf/amf/amfnd/clm.cc >>>>>>>>>> +++ b/osaf/services/saf/amf/amfnd/clm.cc >>>>>>>>>> @@ -37,6 +37,7 @@ >>>>>>>>>> #include "avnd.h" >>>>>>>>>> #include "mds_pvt.h" >>>>>>>>>> #include "nid_api.h" >>>>>>>>>> +#include "osaf_time.h" >>>>>>>>>> >>>>>>>>>> static void clm_node_left(SaClmNodeIdT node_id) >>>>>>>>>> { >>>>>>>>>> @@ -166,7 +167,6 @@ uint32_t >>>> avnd_evt_avd_node_up_evh(AVND_C >>>>>>>>>> info = &evt->info.avd->msg_info.d2n_node_up; >>>>>>>>>> >>>>>>>>>> /*** update this node with the supplied parameters ***/ >>>>>>>>>> - cb->type = info->node_type; >>>>>>>>>> cb->su_failover_max = info->su_failover_max; >>>>>>>>>> cb->su_failover_prob = info->su_failover_prob; >>>>>>>>>> >>>>>>>>>> @@ -249,8 +249,6 @@ done: >>>>>>>>>> return; >>>>>>>>>> } >>>>>>>>>> >>>>>>>>>> -static SaVersionT Version = { 'B', 4, 1 }; >>>>>>>>>> - >>>>>>>>>> static const SaClmCallbacksT_4 callbacks = { >>>>>>>>>> 0, >>>>>>>>>> /*.saClmClusterTrackCallback =*/ clm_track_cb @@ >>>>>>>>>> -263,11 +261,24 @@ SaAisErrorT avnd_clm_init(void) >>>>>>>>>> >>>>>>>>>> TRACE_ENTER(); >>>>>>>>>> avnd_cb->first_time_up = true; >>>>>>>>> [Praveen] Did not get the reason for its removal? >>>>>>>>> Not being updated any where else in the patch. >>>>>>>> [AndersW] Nothing is removed here. The CLM initialization loop >>>>>>>> has been enhanced to handle more error codes (TIMEOUT & >>>> UNAVAILABLE). >>>>>>>>>> - error = saClmInitialize_4(&avnd_cb->clmHandle, &callbacks, >>>>>>>>>> &Version); >>>>>>>>>> - if (SA_AIS_OK != error) { >>>>>>>>>> - LOG_ER("Failed to Initialize with CLM: %u", error); >>>>>>>>>> - goto done; >>>>>>>>>> - } >>>>>>>>>> + for (;;) { >>>>>>>>>> + SaVersionT Version = { 'B', 4, 1 }; >>>>>>>>>> + error = saClmInitialize_4(&avnd_cb->clmHandle, &callbacks, >>>>>>>>>> + &Version); >>>>>>>>>> + if (error == SA_AIS_ERR_TRY_AGAIN || >>>>>>>>>> + error == SA_AIS_ERR_TIMEOUT || >>>>>>>>>> + error == SA_AIS_ERR_UNAVAILABLE) { >>>>>>>>>> + if (error != SA_AIS_ERR_TRY_AGAIN) { >>>>>>>>>> + LOG_WA("saClmInitialize_4 returned %u", >>>>>>>>>> + (unsigned) error); >>>>>>>>>> + } >>>>>>>>>> + osaf_nanosleep(&kHundredMilliseconds); >>>>>>>>>> + continue; >>>>>>>>>> + } >>>>>>>>>> + if (error == SA_AIS_OK) break; >>>>>>>>>> + LOG_ER("Failed to Initialize with CLM: %u", error); >>>>>>>>>> + goto done; >>>>>>>>>> + } >>>>>>>>>> error = saClmSelectionObjectGet(avnd_cb->clmHandle, >>>>>>>>>> &avnd_cb->clm_sel_obj); >>>>>>>>>> if (SA_AIS_OK != error) { >>>>>>>>>> LOG_ER("Failed to get CLM selectionObject: >>>>>>>>>> %u", error); >>>>>>>>>> >>>> --------------------------------------------------------------------- >>>> --------- >>>> Transform Data into Opportunity. >>>> Accelerate data analysis in your applications with Intel Data >>>> Analytics Acceleration Library. >>>> Click to learn more. >>>> http://pubads.g.doubleclick.net/gampad/clk?id=278785471&iu=/4140 >>>> _______________________________________________ >>>> Opensaf-devel mailing list >>>> Opensaf-devel@lists.sourceforge.net >>>> https://lists.sourceforge.net/lists/listinfo/opensaf-devel ------------------------------------------------------------------------------ Transform Data into Opportunity. Accelerate data analysis in your applications with Intel Data Analytics Acceleration Library. Click to learn more. http://pubads.g.doubleclick.net/gampad/clk?id=278785471&iu=/4140 _______________________________________________ Opensaf-devel mailing list Opensaf-devel@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/opensaf-devel
