Hi Thang,
I assume this is the fix we have already discussed with no changes. If that's
the case Ack.
Thanks
Lennart
-----Original Message-----
From: thang.d.nguyen <thang.d.ngu...@dektech.com.au>
Sent: den 26 augusti 2019 12:10
To: Lennart Lund <lennart.l...@ericsson.com>; Vu Minh Nguyen
<vu.m.ngu...@dektech.com.au>
Cc: opensaf-devel@lists.sourceforge.net; Thang Duc Nguyen
<thang.d.ngu...@dektech.com.au>
Subject: [PATCH 1/1] smf: improve ccb handler rules for recovery [#3071]
Retryin ccb handler when ccb aborted to sync.
---
src/smf/smfd/SmfUpgradeAction.cc | 83 +++++-------------------
src/smf/smfd/SmfUpgradeStep.cc | 77 ++++------------------
src/smf/smfd/imm_modify_config/immccb.cc | 32 +++++++--
3 files changed, 57 insertions(+), 135 deletions(-)
diff --git a/src/smf/smfd/SmfUpgradeAction.cc b/src/smf/smfd/SmfUpgradeAction.cc
index b0919c2b8..fca9311c7 100644
--- a/src/smf/smfd/SmfUpgradeAction.cc
+++ b/src/smf/smfd/SmfUpgradeAction.cc
@@ -485,73 +485,26 @@ SaAisErrorT SmfImmCcbAction::execute(SaImmOiHandleT
i_oiHandle,
if (m_operations.size() > 0) {
TRACE("Imm Ccb Action");
- SmfImmUtils immUtil;
- uint32_t retry_count = 0;
- while (1) {
- if (i_rollbackDn != NULL) {
- rollbackCcb =
- new (std::nothrow) SmfRollbackCcb(immRollbackCcbDn, i_oiHandle);
- if (rollbackCcb == NULL) {
- LOG_ER("SmfImmCcbAction::execute failed to create SmfRollbackCcb");
- return SA_AIS_ERR_NO_MEMORY;
- }
+ SmfImmUtils smfImmUtils;
+
+ if (i_rollbackDn != NULL) {
+ rollbackCcb =
+ new (std::nothrow) SmfRollbackCcb(immRollbackCcbDn, i_oiHandle);
+ if (rollbackCcb == NULL) {
+ LOG_ER("SmfImmCcbAction::execute failed to create SmfRollbackCcb");
+ return SA_AIS_ERR_NO_MEMORY;
}
- result = immUtil.doImmOperations(m_operations, rollbackCcb);
- if (((result == SA_AIS_ERR_TIMEOUT) || (result == SA_AIS_ERR_NOT_EXIST))
- && (retry_count <= 6)) {
- int interval = 5; // seconds
- // When IMM aborts a CCB because of synch request from a payload, then
- // the next call of CCBInitialize() will return TRY_AGAIN till the time
- // the synch is complete.
- // There is no direct information available to the OM that can indicate
- // that the CCB or the Adminownerset failed because of an abort and
also
- // there is no notification that can indicate that IMM is ready now.
- // That leaves SMF with the option to correlate error codes returned.
- //
- // Notes on treatment of SA_AIS_ERR_TIMEOUT and SA_AIS_ERR_NOT_EXIST
- // error codes:
- //
- // 1) CCB abort when it is not the first
operation(create/modify/delete)
- // in the CCB
- // and if there is dependency between objects in the CCB:-
- //
- // An abort of a CCB and if the objects(Create/Modify/delete) had
- // some dependency(parent-child) among them, then an API call of
- // AdminOwnerSet() or the CCBCreate/Delete/Modify() on a dependant
- // object will return SA_AIS_ERR_NOT_EXIST, because the CCB aborted.
- //
- // 2) CCB abort when it is a first operation and/or there is no
- // intra-ccb objects-dependency:-
- //
- // When an ongoing CCB is aborted because of a synch request
originating
- // from a payload, then the AdminOwnerSet() or the
- // CCBCreate/Delete/Modify() will return timeout.
-
- ++retry_count;
- LOG_NO("SmfImmCcbAction::execute failed with error: %s",
- saf_error(result));
- LOG_NO("CCB was aborted!?, Retrying: %u", retry_count);
- // Total retry time of 2.5 minutes for a worst case IMM loaded with say
- // < 300k objects. Retry every 25 seconds. i.e. (nanosleep for 5
- // seconds) + (immutil_ccbInitialize will worstcase wait till 20
- // seconds).
- struct timespec sleepTime = {interval, 0};
- osaf_nanosleep(&sleepTime);
- if (rollbackCcb != NULL) {
- delete rollbackCcb;
- rollbackCcb = NULL;
- }
- continue;
- } else if (result != SA_AIS_OK) {
- LOG_ER("SmfImmCcbAction::execute failed, result=%s",
- saf_error(result));
- if (rollbackCcb != NULL) {
- delete rollbackCcb;
- rollbackCcb = NULL;
- }
+ }
+
+ result = smfImmUtils.doImmOperations(m_operations, rollbackCcb);
+ if (result != SA_AIS_OK) {
+ LOG_ER("SmfImmCcbAction::execute failed, result=%s",
+ saf_error(result));
+ if (rollbackCcb != NULL) {
+ delete rollbackCcb;
+ rollbackCcb = NULL;
}
- break;
- }/* End while (1) */
+ }
}
if (rollbackCcb != NULL) {
diff --git a/src/smf/smfd/SmfUpgradeStep.cc b/src/smf/smfd/SmfUpgradeStep.cc
index 60fd08372..a029a63e1 100644
--- a/src/smf/smfd/SmfUpgradeStep.cc
+++ b/src/smf/smfd/SmfUpgradeStep.cc
@@ -692,76 +692,25 @@ SaAisErrorT SmfUpgradeStep::modifyInformationModel() {
if (m_modificationList.size() > 0) {
TRACE("Modifying information model");
- SmfImmUtils immUtil;
- uint32_t retry_count = 0;
-
- while (1) {
- SmfRollbackCcb rollbackCcb(
- modifyRollbackCcbDn,
getProcedure()->getProcThread()->getImmHandle());
-
- rc = immUtil.doImmOperations(m_modificationList, &rollbackCcb);
- if (((rc == SA_AIS_ERR_TIMEOUT) || (rc == SA_AIS_ERR_NOT_EXIST)) &&
- (retry_count <= 6)) {
- int interval = 5; // seconds
- // When IMM aborts a CCB because of synch request from a payload, then
- // the next call of CCBInitialize() will return TRY_AGAIN till the time
- // the synch is complete.
- // There is no direct information available to the OM that can indicate
- // that the CCB or the Adminownerset failed because of an abort and
also
- // there is no notification that can indicate that IMM is ready now.
- // That leaves SMF with the option to correlate error codes returned.
- //
- // Notes on treatment of SA_AIS_ERR_TIMEOUT and SA_AIS_ERR_NOT_EXIST
- // error codes:
- //
- // 1) CCB abort when it is not the first
operation(create/modify/delete)
- // in the CCB
- // and if there is dependency between objects in the CCB:-
- //
- // An abort of a CCB and if the objects(Create/Modify/delete) had
- // some dependency(parent-child) among them, then an API call of
- // AdminOwnerSet() or the CCBCreate/Delete/Modify() on a dependant
- // object will return SA_AIS_ERR_NOT_EXIST, because the CCB aborted.
- //
- // 2) CCB abort when it is a first operation and/or there is no
- // intra-ccb objects-dependency:-
- //
- // When an ongoing CCB is aborted because of a synch request
originating
- // from a payload, then the AdminOwnerSet() or the
- // CCBCreate/Delete/Modify() will return timeout.
-
- ++retry_count;
- LOG_NO("SmfUpgradeStep modify IMM failed with error: %s",
- saf_error(rc));
- LOG_NO("CCB was aborted!?, Retrying: %u", retry_count);
- // Total retry time of 2.5 minutes for a worst case IMM loaded with say
- // < 300k objects. Retry every 25 seconds. i.e. (nanosleep for 5
- // seconds) + (immutil_ccbInitialize will worstcase wait till 20
- // seconds).
- struct timespec sleepTime = {interval, 0};
- osaf_nanosleep(&sleepTime);
- // Note: Alternatively Make rollbackCcb unique by adding a method for
- // this to the rollbackCcb.
- continue;
- } else if (rc != SA_AIS_OK) {
- LOG_NO("Giving up, SmfUpgradeStep modify IMM failed, rc=%s",
- saf_error(rc));
- return rc;
- } else { /* Things went fine */
+ SmfImmUtils smfimmUtils;
- if ((rc = rollbackCcb.execute()) != SA_AIS_OK) {
- LOG_NO("SmfUpgradeStep failed to store rollback CCB, rc=%s",
- saf_error(rc));
- return rc;
- }
- break;
+ SmfRollbackCcb rollbackCcb(
+ modifyRollbackCcbDn,
+ getProcedure()->getProcThread()->getImmHandle());
+
+ rc = smfimmUtils.doImmOperations(m_modificationList, &rollbackCcb);
+ if (rc == SA_AIS_OK) {
+ if ((rc = rollbackCcb.execute()) != SA_AIS_OK) {
+ LOG_NO("SmfUpgradeStep failed to store rollback CCB, rc=%s",
+ saf_error(rc));
}
- } /* End while (1) */
+ } else {
+ LOG_NO("SmfUpgradeStep::modifyInformationModel: Fail, %s",
saf_error(rc));
+ }
} else {
TRACE("Nothing to modify in information model");
}
- return SA_AIS_OK;
+ return rc;
}
//------------------------------------------------------------------------------
diff --git a/src/smf/smfd/imm_modify_config/immccb.cc
b/src/smf/smfd/imm_modify_config/immccb.cc
index cc62911db..c8cdac2d3 100644
--- a/src/smf/smfd/imm_modify_config/immccb.cc
+++ b/src/smf/smfd/imm_modify_config/immccb.cc
@@ -220,11 +220,15 @@ int ModelModification::CreateObjectManager() {
bool return_state = imm_om_handle_->InitializeHandle();
if (return_state == false) {
- // No recovery is possible
- LOG_NO("%s: OM-handle, RestoreHandle(), Fail", __FUNCTION__);
- recovery_info = kFail;
- api_name_ = "saImmOmInitialize";
- ais_error_ = imm_om_handle_->ais_error();
+ SaAisErrorT ais_error = imm_om_handle_->ais_error();
+ if (ais_error == SA_AIS_ERR_TRY_AGAIN) {
+ recovery_info = kRestartOm;
+ } else {
+ // No recovery is possible
+ LOG_NO("%s: OM-handle, RestoreHandle(), Fail", __FUNCTION__);
+ recovery_info = kFail;
+ api_name_ = "saImmOmInitialize";
+ }
} else {
recovery_info = kContinue;
}
@@ -418,6 +422,14 @@ int ModelModification::AdminOwnerSet(const
std::vector<std::string>& objects,
return recovery_info;
}
+static bool has_object(const std::vector<CreateDescriptor>& create_descriptors,
+ const std::string& object) {
+ for (const auto& desc : create_descriptors) {
+ if (object == desc.parent_name) return true;
+ }
+ return false;
+}
+
// Add create requests for all objects to be created // Set admin ownership
for parent to all objects to be created with scope // SA_IMM_ONE.
@@ -440,9 +452,17 @@ int ModelModification::AddCreates(const
std::vector<CreateDescriptor>&
// Become admin owner of parent if there is a parent. If no parent the
// IMM object will be created as a root object
std::vector<std::string> imm_objects;
- imm_objects.push_back(create_descriptor.parent_name);
+ std::string parent = create_descriptor.parent_name;
+ imm_objects.push_back(parent);
recovery_info = AdminOwnerSet(imm_objects, SA_IMM_ONE);
if (recovery_info == kFail) {
+ // data for this ccb is valid but it is aborted by imm to start
+ // IMM sync, therefore marking this ccb is restartable.
+ if (ais_error_ == SA_AIS_ERR_NOT_EXIST &&
+ has_object(create_descriptors, parent)) {
+ recovery_info = kRestartOm;
+ break;
+ }
LOG_NO("%s: AdminOwnerSet() Fail", __FUNCTION__);
break;
} else if (recovery_info == kRestartOm) {
--
2.17.1
_______________________________________________
Opensaf-devel mailing list
Opensaf-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/opensaf-devel
