Detailed Status Ceph in PG

　　Creating
　　
　　Peering
　　
　　Activating
　　
　　the Active
　　
　　Backfilling
　　
　　Backfill-toofull
　　
　　Backfill-the wait
　　
　　is in Incomplete state
　　
　　Inconsistent
　　
　　Peered
　　
　　Recovering
　　
　　Recovering-the wait
　　
　　the REMAPPED
　　
　　The Scrubbing
　　
　　unactive
　　
　　Unclean
　　
　　Stale
　　
　　Undersize
　　
　　Down
　　
　　Creating
　　
　　Meaning: PG are creating
　　
　　cause: to create pool of time, while creating pg conducted according to the specified pg number state, the normal state of the
　　
　　consequences: no
　　
　　solution: no need to solve one of the normal state
　　
　　Peering
　　
　　meaning: interconnection between PG, agree on one of the objects and metadata states
　　
　　the cause: when pg is creating, will be interconnected storage to agree on which the objects and metadata group copy status between opposing normalized through the OSD.
　　
　　Consequences: No
　　
　　solution: no need to solve one of the normal state
　　
　　Activating
　　
　　Meaning: pg after completing the peering process, previous results will be cured, wait for all pg sync, try to enter the active state
　　
　　the cause of: pg ready to enter the active state before
　　
　　the consequences: if long-term card in this state, will not be affected by the PG reading and writing, thereby affecting the entire pool availability
　　
　　solutions: PG stopped where all OSD
　　
　　perform data backup pg ceph-object-tool
　　
　　with ceph-object-tool deletes empty on the primary PG pg (Do not manually delete)
　　
　　used again ceph- object-tool to import data
　　
　　manually to the directory pg Fu ceph permissions
　　
　　final restart osd to
　　
　　Active
　　
　　meaning: pg active state, can read and write
　　
　　the cause of: normal
　　
　　consequences: None
　　
　　solution: no need to solve one of the normal state
　　
　　Backfilling
　　
　　meaning: status backfilling
　　
　　causes of: osd this is usually due to the off-line (no more than 5 minutes heartbeat response), ceph find new osd to replace the total amount of data copy performed.
　　
　　Consequences: the emergence of this state are generally determined to hang or offline osd a
　　
　　solution: ceph will auto-complete data backfill In most cases, if you can not complete the backfill, backfill-toofull will enter the state
　　
　　Backfill-toofull
　　
　　Backfilling suspended state: Meaning
　　
　　cause reason: because the cause is usually not large enough to backfill missing osd osd
　　
　　Consequences: Causes pool can not be written, read and write stuck.
　　
　　Solution: need to check osd capacity, whether there are serious imbalances, the excess data osd manual evacuation (reweight), if the cluster nearful phenomenon, should as soon as possible the physical expansion of
　　
　　the emergency expansion mode (temporary solution, the best way is extended osd number and capacity)
　　
　　to suspend write osd:
　　
　　Ceph osd pAUSE
　　
　　notification mon osd modifications and full threshold
　　
　　Ceph mon * injectargs Tell "--mon-osd full-ratio-0.96."
　　
　　Ceph osd * injectargs Tell "--mon-osd. -full-ratio 0.96 "
　　
　　notification PG modified full threshold:
　　
　　Ceph PG set_full_ratio 0.96
　　
　　releasing write prohibition osd:
　　
　　Ceph osd Unpause
　　
　　backfill the wait-
　　
　　meaning: PG are waiting to begin backfilling operation.
　　
　　The cause of: OSD off cause (not personally capture the state, probably too fast did not see)
　　
　　consequences: The next theory in terms of pg enter data backfill backfilling state
　　
　　solution: normal backfill must pass through state, no special attention
　　
　　Incomplete
　　
　　meanings: peering process found not agree on the state of the data
　　
　　causes: pg authority selected when the log, the log can not complete the authority, the authority of the log or log is completed and the local contrast is not normal logic
　　
　　Consequences: usually leads pg can not be created, stuck in creating + incomplete state, which led pool can not be used
　　
　　Solution: First make sure osd_allow_recovery_below_min_size is true, as well as the number of copies is reasonable, whether the number of select osd configuration crushmap consistent with the pool, if both normal, try the following recovery process
　　
　　stopped all incomplete corresponding to each of the PG osd
　　
　　use ceph-object-tool to mark complete osd were
　　
　　then restart osd
　　
　　inconsistent
　　
　　meanings: in fact inconsistent copy of the data mean
　　
　　the cause of: a copy of the data is unknown the reason is missing
　　
　　consequences: a copy of the data inconsistency resulting in reduced safety
　　
　　solution: use ceph pg repair tools for data recovery, under normal circumstances can be restored to normal, if you can not recover
　　
　　the osd of osd_max_scrubs three copies are to emphasize large, that use ceph again pg repair tools for data recovery, then finally transferred back to osd_max_scrubs 1
　　
　　Peered
　　
　　meanings: search, referring to the PG can not find enough copies to read and write operations (even min_size are unable to meet the love Under)
　　
　　the cause of: multiple osd hang up, causing the number of copies of the currently active osd <min_size, read and write functions lock
　　
　　consequences: pg can not be used, not even for routine pool io
　　
　　Solution: cluster health status, osd hang over five minutes will automatically remapped to repair the state, the state wants a quick fix method is twofold:
　　
　　1 try to start a copy osd, rejoins the cluster, peered will automatically disappear
　　
　　osd 2 initiative out of loss of associated, ceph will automatically enter the repair status
　　
　　Recovering
　　
　　Meaning: Restoring the
　　
　　causes of: when an OSD hung up (down), within which the group will return home behind a copy of the other co-location group; when this OSD rebirth (up), co-location set of content must be updated to the current status;
　　
　　consequences: the recovery is not always the little things, because of a hardware failure may be implicated in multiple OSD. For example, a cabinet or room network switches fail, which can lead to the current state of the OSD on multiple hosts behind the cluster, each OSD after recovery must be restored.
　　
　　Solution: Cluster emergence of this state, indicating that PG is automatically restored, wait for it to recover complete enough.
　　
　　Recovering-wait
　　
　　Meaning: Wait for Recovery resource reservation
　　
　　causes of: PG awaiting restoration.
　　
　　Consequences: pg theory in terms of recovering state will enter the data recovery
　　
　　solutions: restoring normal state.
　　
　　Remapped
　　
　　meanings: remapping state
　　
　　causes of: when set inside the PG combination Acting changed, a new data set from the old to the migration sets. This time may be relatively long, the main OSD new collection before the completion of the migration can not respond to the request. So the new master OSD will be asked to continue to serve the guidance of former team OSD PG migration is complete. Once the data migration is complete, the new primary will take effect OSD accept the request.
　　
　　Consequences: If you can not re-mapping, data can not be migrated, you may lose data.
　　
　　Solution: hang in the OSD or OSD on PG PG reassigned osd number belongs Crush algorithm according to the time of expansion. And PG Remap will go to other OSD.
　　
　　When Remapped state, PG and current Acting Set Up Set inconsistent.
　　
　　IO client can read and write properly.
　　
　　Scrubbing
　　
　　meaning: clean up the
　　
　　causes of: pg inconsistency check is being done.
　　
　　Consequences: IO will cause performance degradation
　　
　　solution: according to the actual needs of the environment, turn the feature off or reduce the frequency of self-test.
　　
　　Unactive
　　
　　meanings: inactive state, PG read and write requests can not be processed
　　
　　due to reasons: PG for a long time does not appear as acitve state (read or write request can not be executed), PG can not read or write,
　　
　　consequences: PG can not perform read and write
　　
　　solution : wait OSD update data to the latest backup status
　　
　　Unclean
　　
　　meaning: uncleaned state, PG can not recover from a failure
　　
　　caused by reasons: return home the group number of copies some object does not reach the desired number of times, they should be in recovery;
　　
　　the consequences : data security decreased
　　
　　solution: usually have to perform the recovery operation
　　
　　Stale
　　
　　meaning: to refresh the state, pg has not been any updates osd
　　
　　caused by reasons: it may be caused by osd hang, follow peering status appear together under normal circumstances
　　
　　Analogue: manual stopped a osd, systemctl stop ceph-osd @ 0, see ceph -s will find that in a short time (peering before), pg will enter stale + clean + active status of special
　　
　　effects: warning signs, often represents osd abnormality occurs off network or a node.
　　
　　Solution: In general just have to wait peering can be completed.
　　
　　Undersized
　　
　　Meaning: the number of copies is too small
　　
　　causes of: the number of copies is less than the number of copies of the PG configured storage pool. Usually due to a osd down the service, when this state.
　　
　　Consequences: reduce data availability
　　
　　solutions: PG adjustment pool where the number of copies osd default min size pool = 1, no adjustment is recommended. Osd up and other services like
　　
　　# delete news of stop words
　　
　　DEF drop_stopwords (Contents, stopwords):
　　
　　contents_clean = [] # deleted after the news
　　
　　all_words = [www.qjljdgt.cn] # word cloud structure data used
　　
　　for line Contents in:
　　
　　line_clean = [www.zheshengyule.com]
　　
　　for in Word Line:
　　
　　IF stopwords in Word:
　　
　　Continue
　　
　　line_clean.append (Word)
　　
　　all_words.append (STR (Word))
　　
　　contents_clean.append (line_clean)
　　
　　return contents_clean, all_words
　　
　　Contents = www.tyyLeapp.com df_content.content_S.values.tolist ()
　　
　　stopwords = stopwords.stopword.values.tolist ()
　　
　　after the # delete stop words get news and word cloud data
　　
　　contents_clean, all_words = drop_stopwords (Contents, stopwords)
　　
　　# df_content.content_S.isin (stopwords.stopword)
　　
　　# = df_content df_content [www.jintianxuesha.com ~ df_content.content_S.isin (stopwords.stopword)]
　　
　　# df_content.head ()
　　
　　View news content after deleting the stop word
　　
　　df_content = pd.DataFrame ({www.mxdpt.cn'contents_clean ': contents_clean})
　　
　　df_content.head ()
　　
　　can be seen from the results, the comparison of the data for improving the quality of the above data a lot.
　　
　　Look at all the words appear, that is, delete all_words after the stop words.
　　
　　www.sanguoyoux.cn = pd.DataFrame df_all_words ({ 'all_words': all_words})
　　
　　df_all_words.head ()
　　
　　results:
　　
　　statistical word frequency all_words each word, the word frequency statistics for convenience and showing the back of a word cloud.
　　
　　numpy Import
　　
　　# packet Statistical
　　
　　words_count = df_all_words.groupby (by = [ 'all_words']) [ 'all_words'] AGG ({ 'count': numpy.size}).
　　
　　# The count sort
　　
　　. words_count = words_count.reset_index () sort_values (by = [ 'COUNT'], Ascending = False)
　　
　　words_count.head (www.pingguoyul.cn)
　　
　　meaning: failure
　　
　　caused by reasons: authoritative copy of the OSD downtime return home group, you must wait for it to boot, or is marked as lost to continue
　　
　　consequences: this time the PG can not provide clients read and write IO, IO hang rammed live
　　
　　solution: OSD up service.