Below is the list of changes that have just been committed into a local
5.1 repository of mskold. When mskold does a push these changes
will be propagated to the main repository and, within 24 hours after the
push, to the public repository.
For information on how to access the public repository
see http://dev.mysql.com/doc/mysql/en/installing-source-tree.html
ChangeSet@stripped, 2008-02-20 13:52:06+01:00, mskold@stripped +5 -0
WL#4083 Resource shortage handling in event mechanism: Added support for handling of buffer overflow in Suma during node failure and disconnect of lagging subscribers (subscribers that do not acknowledge GCP's fast enough), redid implementation for micro-GCPs
storage/ndb/include/mgmapi/mgmapi_config_parameters.h@stripped, 2008-02-20 13:51:58+01:00, mskold@stripped +1 -1
WL#4083 Resource shortage handling in event mechanism: Added support for handling of buffer overflow in Suma during node failure and disconnect of lagging subscribers (subscribers that do not acknowledge GCP's fast enough), redid implementation for micro-GCPs
storage/ndb/src/common/debugger/EventLogger.cpp@stripped, 2008-02-20 13:51:58+01:00, mskold@stripped +3 -3
WL#4083 Resource shortage handling in event mechanism: Added support for handling of buffer overflow in Suma during node failure and disconnect of lagging subscribers (subscribers that do not acknowledge GCP's fast enough), redid implementation for micro-GCPs
storage/ndb/src/kernel/blocks/suma/Suma.cpp@stripped, 2008-02-20 13:51:58+01:00, mskold@stripped +37 -54
WL#4083 Resource shortage handling in event mechanism: Added support for handling of buffer overflow in Suma during node failure and disconnect of lagging subscribers (subscribers that do not acknowledge GCP's fast enough), redid implementation for micro-GCPs
storage/ndb/src/kernel/blocks/suma/Suma.hpp@stripped, 2008-02-20 13:51:58+01:00, mskold@stripped +5 -4
WL#4083 Resource shortage handling in event mechanism: Added support for handling of buffer overflow in Suma during node failure and disconnect of lagging subscribers (subscribers that do not acknowledge GCP's fast enough), redid implementation for micro-GCPs
storage/ndb/src/mgmsrv/ConfigInfo.cpp@stripped, 2008-02-20 13:51:58+01:00, mskold@stripped +1 -1
WL#4083 Resource shortage handling in event mechanism: Added support for handling of buffer overflow in Suma during node failure and disconnect of lagging subscribers (subscribers that do not acknowledge GCP's fast enough), redid implementation for micro-GCPs
diff -Nrup a/storage/ndb/include/mgmapi/mgmapi_config_parameters.h b/storage/ndb/include/mgmapi/mgmapi_config_parameters.h
--- a/storage/ndb/include/mgmapi/mgmapi_config_parameters.h 2008-02-11 14:24:16 +01:00
+++ b/storage/ndb/include/mgmapi/mgmapi_config_parameters.h 2008-02-20 13:51:58 +01:00
@@ -123,7 +123,7 @@
#define CFG_DB_MICRO_GCP_INTERVAL 170 /* micro gcp */
#define CFG_DB_MICRO_GCP_TIMEOUT 171
-#define CFG_DB_MAX_BUFFERED_GCP 180 /* subscriptions */
+#define CFG_DB_MAX_BUFFERED_EPOCHS 182 /* subscriptions */
#define CFG_DB_SGA 198 /* super pool mem */
#define CFG_DB_DATA_MEM_2 199 /* used in special build in 5.1 */
diff -Nrup a/storage/ndb/src/common/debugger/EventLogger.cpp b/storage/ndb/src/common/debugger/EventLogger.cpp
--- a/storage/ndb/src/common/debugger/EventLogger.cpp 2008-02-13 11:39:47 +01:00
+++ b/storage/ndb/src/common/debugger/EventLogger.cpp 2008-02-20 13:51:58 +01:00
@@ -996,10 +996,10 @@ void getTextSubscriptionStatus(QQQQ)
case(1): // SubscriptionStatus::DISCONNECTED
BaseString::snprintf(m_text, m_text_len,
"Disconnecting node %u because it has "
- "exceeded MaxBufferedEpochs (%llu > %u), gci %llu",
+ "exceeded MaxBufferedEpochs (%u > %u), gci %llu",
theData[2],
- make_uint64(theData[5], theData[6]),
- theData[7],
+ theData[5],
+ theData[6],
make_uint64(theData[3], theData[4]));
break;
case(2): // SubscriptionStatus::INCONSISTENT
diff -Nrup a/storage/ndb/src/kernel/blocks/suma/Suma.cpp b/storage/ndb/src/kernel/blocks/suma/Suma.cpp
--- a/storage/ndb/src/kernel/blocks/suma/Suma.cpp 2008-02-19 09:10:33 +01:00
+++ b/storage/ndb/src/kernel/blocks/suma/Suma.cpp 2008-02-20 13:51:58 +01:00
@@ -165,13 +165,13 @@ Suma::execREAD_CONFIG_REQ(Signal* signal
ndbrequire(p != 0);
// SumaParticipant
- Uint32 noTables, noAttrs, maxBufferedGcp;
+ Uint32 noTables, noAttrs, maxBufferedEpochs;
ndb_mgm_get_int_parameter(p, CFG_DB_NO_TABLES,
&noTables);
ndb_mgm_get_int_parameter(p, CFG_DB_NO_ATTRIBUTES,
&noAttrs);
- ndb_mgm_get_int_parameter(p, CFG_DB_MAX_BUFFERED_GCP,
- &maxBufferedGcp);
+ ndb_mgm_get_int_parameter(p, CFG_DB_MAX_BUFFERED_EPOCHS,
+ &maxBufferedEpochs);
c_tablePool.setSize(noTables);
c_tables.setSize(noTables);
@@ -183,7 +183,7 @@ Suma::execREAD_CONFIG_REQ(Signal* signal
c_syncPool.setSize(2);
c_dataBufferPool.setSize(noAttrs);
- c_maxBufferedGcp = maxBufferedGcp;
+ c_maxBufferedEpochs = maxBufferedEpochs;
// Calculate needed gcp pool as 10 records + the ones needed
// during a possible api timeout
@@ -783,7 +783,7 @@ Suma::execNODE_FAILREP(Signal* signal){
{
ndbout_c("Inserting API_FAILREQ node: %u", node);
signal->theData[0] = node;
- EXECUTE_DIRECT(QMGR, GSN_API_FAILREQ, signal, 1);
+ sendSignal(QMGR_REF, GSN_API_FAILREQ, signal, 1, JBA);
}
}
@@ -3561,73 +3561,56 @@ Suma::execFIRE_TRIG_ORD(Signal* signal)
}
void
-Suma::checkMaxBufferedGCP(Signal *signal)
+Suma::checkMaxBufferedEpochs(Signal *signal)
{
/*
* Check if any subscribers are exceeding the MaxBufferedEpochs
*/
+ Ptr<Gcp_record> gcp;
jamEntry();
if (c_gcp_list.isEmpty())
{
jam();
return;
}
- Ptr<Gcp_record> gcp;
c_gcp_list.first(gcp);
if (ERROR_INSERTED(13035))
{
jam();
CLEAR_ERROR_INSERT_VALUE;
ndbout_c("Simulating exceeding the MaxBufferedEpochs %u(%llu,%llu,%llu)",
- c_maxBufferedGcp, m_max_seen_gci,
+ c_maxBufferedEpochs, m_max_seen_gci,
m_last_complete_gci, gcp.p->m_gci);
- c_maxBufferedGcp = 1;
}
- if (m_max_seen_gci - gcp.p->m_gci >= (Uint64) c_maxBufferedGcp)
+ else if (c_gcp_list.count() < c_maxBufferedEpochs)
{
- NodeBitmask subs = c_subscriber_nodes;
- jam();
- // Disconnect lagging subscribers
- for(; !gcp.isNull(); c_gcp_list.next(gcp))
- {
- Uint64 lag = m_max_seen_gci - gcp.p->m_gci;
- jam();
- if (lag >= (Uint64) c_maxBufferedGcp)
- {
- jam();
- for(Uint32 nodeId = 0; nodeId < MAX_NODES; nodeId++)
- {
- if (subs.get(nodeId))
- {
- jam();
- subs.clear(nodeId);
- // Disconnecting node
- signal->theData[0] = NDB_LE_SubscriptionStatus;
- signal->theData[1] = 1; // DISCONNECTED;
- signal->theData[2] = nodeId;
- signal->theData[3] = (Uint32) gcp.p->m_gci;
- signal->theData[4] = (Uint32) (gcp.p->m_gci >> 32);
- signal->theData[5] = (Uint32) lag;
- signal->theData[6] = (Uint32) (lag >> 32);
- signal->theData[7] = c_maxBufferedGcp;
- sendSignal(CMVMI_REF, GSN_EVENT_REP, signal, 8, JBB);
-
- /**
- * Force API_FAILREQ
- */
- signal->theData[0] = nodeId;
- EXECUTE_DIRECT(QMGR, GSN_API_FAILREQ, signal, 1);
- }
- }
- }
- else
- {
- /*
- * We have found a newer gci that still is
- * allowed to be buffered
- */
- break;
- }
+ return;
+ }
+ NodeBitmask subs = gcp.p->m_subscribers;
+ jam();
+ // Disconnect lagging subscribers waiting for oldest epoch
+ ndbout_c("Found lagging epoch %llu", gcp.p->m_gci);
+ for(Uint32 nodeId = 0; nodeId < MAX_NODES; nodeId++)
+ {
+ if (subs.get(nodeId))
+ {
+ jam();
+ subs.clear(nodeId);
+ // Disconnecting node
+ signal->theData[0] = NDB_LE_SubscriptionStatus;
+ signal->theData[1] = 1; // DISCONNECTED;
+ signal->theData[2] = nodeId;
+ signal->theData[3] = (Uint32) gcp.p->m_gci;
+ signal->theData[4] = (Uint32) (gcp.p->m_gci >> 32);
+ signal->theData[5] = (Uint32) c_gcp_list.count();
+ signal->theData[6] = c_maxBufferedEpochs;
+ sendSignal(CMVMI_REF, GSN_EVENT_REP, signal, 8, JBB);
+
+ /**
+ * Force API_FAILREQ
+ */
+ signal->theData[0] = nodeId;
+ sendSignal(QMGR_REF, GSN_API_FAILREQ, signal, 1, JBA);
}
}
}
@@ -3671,7 +3654,7 @@ Suma::execSUB_GCP_COMPLETE_REP(Signal* s
#endif
m_last_complete_gci = gci;
- //checkMaxBufferedGCP(signal);
+ checkMaxBufferedEpochs(signal);
m_max_seen_gci = (gci > m_max_seen_gci ? gci : m_max_seen_gci);
/**
diff -Nrup a/storage/ndb/src/kernel/blocks/suma/Suma.hpp b/storage/ndb/src/kernel/blocks/suma/Suma.hpp
--- a/storage/ndb/src/kernel/blocks/suma/Suma.hpp 2008-02-11 14:24:16 +01:00
+++ b/storage/ndb/src/kernel/blocks/suma/Suma.hpp 2008-02-20 13:51:58 +01:00
@@ -23,7 +23,7 @@
#include <SLList.hpp>
#include <DLList.hpp>
-#include <DLFifoList.hpp>
+#include <DLCFifoList.hpp>
#include <KeyTable.hpp>
#include <DataBuffer.hpp>
#include <SignalCounter.hpp>
@@ -349,7 +349,7 @@ public:
ArrayPool<SyncRecord> c_syncPool;
DataBuffer<15>::DataBufferPool c_dataBufferPool;
- Uint32 c_maxBufferedGcp;
+ Uint32 c_maxBufferedEpochs;
NodeBitmask c_failedApiNodes;
@@ -602,7 +602,8 @@ private:
Uint64 get_current_gci(Signal*);
- void checkMaxBufferedGCP(Signal *signal);
+ void checkMaxBufferedEpochs
+(Signal *signal);
Uint64 m_max_seen_gci; // FIRE_TRIG_ORD
Uint64 m_max_sent_gci; // FIRE_TRIG_ORD -> send
@@ -622,7 +623,7 @@ private:
Uint32 prevList;
};
ArrayPool<Gcp_record> c_gcp_pool;
- DLFifoList<Gcp_record> c_gcp_list;
+ DLCFifoList<Gcp_record> c_gcp_list;
struct Page_chunk
{
diff -Nrup a/storage/ndb/src/mgmsrv/ConfigInfo.cpp b/storage/ndb/src/mgmsrv/ConfigInfo.cpp
--- a/storage/ndb/src/mgmsrv/ConfigInfo.cpp 2008-02-11 15:05:02 +01:00
+++ b/storage/ndb/src/mgmsrv/ConfigInfo.cpp 2008-02-20 13:51:58 +01:00
@@ -917,7 +917,7 @@ const ConfigInfo::ParamInfo ConfigInfo::
"32000" },
{
- CFG_DB_MAX_BUFFERED_GCP,
+ CFG_DB_MAX_BUFFERED_EPOCHS,
"MaxBufferedEpochs",
DB_TOKEN,
"Allowed numbered of epochs that a subscribing node can lag behind (unprocessed epochs). Exceeding will cause lagging subscribers to be disconnected.",
| Thread |
|---|
| • bk commit into 5.1 tree (mskold:1.2531) | Martin Skold | 20 Feb |
