Re: [Gluster-users] question about info and info.tmp
On Fri, Nov 25, 2016 at 1:14 PM, songxinwrote: > Hi Atin, > It seems that this workaround should be done by manual. > Is that right? > And even the files in bricks/* may be empty too. > Yes, that's right > > Do you have a workaround, which is implemented in glusterfs code? > Workaround is by nature manual and anything to be done through code should be considered as fix not work around :) > > Thanks, > Xin > > > > > > 在 2016-11-25 15:36:29,"Atin Mukherjee" 写道: > > > > On Fri, Nov 25, 2016 at 12:06 PM, songxin wrote: > >> Hi Atin, >> Do you mean that you have the workaround applicable now? >> Or it will take time to design the workaround? >> >> If you have workaround now, could you share it to me ? >> > > If you end up in having a 0 byte info file you'd need to copy the same > info file from other node and put it there and restart glusterd. > > >> >> Thanks, >> Xin, >> >> >> >> >> >> 在 2016-11-24 19:12:07,"Atin Mukherjee" 写道: >> >> Xin - I appreciate your patience. I'd need some more time to pick this >> item up from my backlog. I believe we have a workaround applicable here too. >> >> On Thu, 24 Nov 2016 at 14:24, songxin wrote: >> >>> >>> >>> >>> Hi Atin, >>> Actually, the glusterfs is used in my project. >>> And our test team find this issue. >>> So I want to make sure that whether you plan to fix it. >>> if you have plan I will wait you because your method shoud be better >>> than mine. >>> >>> Thanks, >>> Xin >>> >>> >>> 在 2016-11-21 10:00:36,"Atin Mukherjee" 写道: >>> >>> Hi Xin, >>> >>> I've not got a chance to look into it yet. delete stale volume function >>> is in place to take care of wiping off volume configuration data which has >>> been deleted from the cluster. However we need to revisit this code to see >>> if this function is anymore needed given we recently added a validation to >>> fail delete request if one of the glusterd is down. I'll get back to you on >>> this. >>> >>> On Mon, 21 Nov 2016 at 07:24, songxin wrote: >>> >>> Hi Atin, >>> Thank you for your support. >>> >>> And any conclusions about this issue? >>> >>> Thanks, >>> Xin >>> >>> >>> >>> >>> >>> 在 2016-11-16 20:59:05,"Atin Mukherjee" 写道: >>> >>> >>> >>> On Tue, Nov 15, 2016 at 1:53 PM, songxin wrote: >>> >>> ok, thank you. >>> >>> >>> >>> >>> 在 2016-11-15 16:12:34,"Atin Mukherjee" 写道: >>> >>> >>> >>> On Tue, Nov 15, 2016 at 12:47 PM, songxin wrote: >>> >>> >>> Hi Atin, >>> >>> I think the root cause is in the function glusterd_import_friend_volume >>> as below. >>> >>> int32_t >>> glusterd_import_friend_volume (dict_t *peer_data, size_t count) >>> { >>> ... >>> ret = glusterd_volinfo_find (new_volinfo->volname, >>> _volinfo); >>> if (0 == ret) { >>> (void) gd_check_and_update_rebalance_info (old_volinfo, >>>n >>> ew_volinfo); >>> (void) glusterd_delete_stale_volume (old_volinfo, >>> new_volinfo); >>> } >>> ... >>> ret = glusterd_store_volinfo (new_volinfo, >>> GLUSTERD_VOLINFO_VER_AC_NONE); >>> if (ret) { >>> gf_msg (this->name, GF_LOG_ERROR, 0, >>> GD_MSG_VOLINFO_STORE_FAIL, "Failed to store " >>> "volinfo for volume %s", new_volinfo->volname); >>> goto out; >>> } >>> ... >>> } >>> >>> glusterd_delete_stale_volume will remove the info and bricks/* and the >>> glusterd_store_volinfo will create the new one. >>> But if glusterd is killed before rename the info will is empty. >>> >>> And glusterd will start failed because the infois empty in the next time >>> you start the glusterd. >>> >>> Any idea, Atin? >>> >>> >>> Give me some time, will check it out, but reading at this analysis looks >>> very well possible if a volume is changed when the glusterd was done on >>> node a and when the same comes up during peer handshake we update the >>> volinfo and during that time glusterd goes down once again. I'll confirm it >>> by tomorrow. >>> >>> >>> I checked the code and it does look like you have got the right RCA for >>> the issue which you simulated through those two scripts. However this can >>> happen even when you try to create a fresh volume and while glusterd tries >>> to write the content into the store and goes down before renaming the >>> info.tmp file you get into the same situation. >>> >>> I'd really need to think through if this can be fixed. Suggestions are >>> always appreciated. >>> >>> >>> >>> >>> BTW, excellent work Xin! >>> >>> >>> Thanks, >>> Xin >>> >>> >>> 在 2016-11-15 12:07:05,"Atin Mukherjee" 写道: >>> >>> >>> >>> On Tue, Nov 15, 2016 at 8:58 AM, songxin wrote: >>> >>> Hi Atin, >>> I have some
Re: [Gluster-users] question about info and info.tmp
Hi Atin, It seems that this workaround should be done by manual. Is that right? And even the files in bricks/* may be empty too. Do you have a workaround, which is implemented in glusterfs code? Thanks, Xin 在 2016-11-25 15:36:29,"Atin Mukherjee"写道: On Fri, Nov 25, 2016 at 12:06 PM, songxin wrote: Hi Atin, Do you mean that you have the workaround applicable now? Or it will take time to design the workaround? If you have workaround now, could you share it to me ? If you end up in having a 0 byte info file you'd need to copy the same info file from other node and put it there and restart glusterd. Thanks, Xin, 在 2016-11-24 19:12:07,"Atin Mukherjee" 写道: Xin - I appreciate your patience. I'd need some more time to pick this item up from my backlog. I believe we have a workaround applicable here too. On Thu, 24 Nov 2016 at 14:24, songxin wrote: Hi Atin, Actually, the glusterfs is used in my project. And our test team find this issue. So I want to make sure that whether you plan to fix it. if you have plan I will wait you because your method shoud be better than mine. Thanks, Xin 在 2016-11-21 10:00:36,"Atin Mukherjee" 写道: Hi Xin, I've not got a chance to look into it yet. delete stale volume function is in place to take care of wiping off volume configuration data which has been deleted from the cluster. However we need to revisit this code to see if this function is anymore needed given we recently added a validation to fail delete request if one of the glusterd is down. I'll get back to you on this. On Mon, 21 Nov 2016 at 07:24, songxin wrote: Hi Atin, Thank you for your support. And any conclusions about this issue? Thanks, Xin 在 2016-11-16 20:59:05,"Atin Mukherjee" 写道: On Tue, Nov 15, 2016 at 1:53 PM, songxin wrote: ok, thank you. 在 2016-11-15 16:12:34,"Atin Mukherjee" 写道: On Tue, Nov 15, 2016 at 12:47 PM, songxin wrote: Hi Atin, I think the root cause is in the function glusterd_import_friend_volume as below. int32_t glusterd_import_friend_volume (dict_t *peer_data, size_t count) { ... ret = glusterd_volinfo_find (new_volinfo->volname, _volinfo); if (0 == ret) { (void) gd_check_and_update_rebalance_info (old_volinfo, new_volinfo); (void) glusterd_delete_stale_volume (old_volinfo, new_volinfo); } ... ret = glusterd_store_volinfo (new_volinfo, GLUSTERD_VOLINFO_VER_AC_NONE); if (ret) { gf_msg (this->name, GF_LOG_ERROR, 0, GD_MSG_VOLINFO_STORE_FAIL, "Failed to store " "volinfo for volume %s", new_volinfo->volname); goto out; } ... } glusterd_delete_stale_volume will remove the info and bricks/* and the glusterd_store_volinfo will create the new one. But if glusterd is killed before rename the info will is empty. And glusterd will start failed because the infois empty in the next time you start the glusterd. Any idea, Atin? Give me some time, will check it out, but reading at this analysis looks very well possible if a volume is changed when the glusterd was done on node a and when the same comes up during peer handshake we update the volinfo and during that time glusterd goes down once again. I'll confirm it by tomorrow. I checked the code and it does look like you have got the right RCA for the issue which you simulated through those two scripts. However this can happen even when you try to create a fresh volume and while glusterd tries to write the content into the store and goes down before renaming the info.tmp file you get into the same situation. I'd really need to think through if this can be fixed. Suggestions are always appreciated. BTW, excellent work Xin! Thanks, Xin 在 2016-11-15 12:07:05,"Atin Mukherjee" 写道: On Tue, Nov 15, 2016 at 8:58 AM, songxin wrote: Hi Atin, I have some clues about this issue. I could reproduce this issue use the scrip that mentioned in https://bugzilla.redhat.com/show_bug.cgi?id=1308487 . I really appreciate your help in trying to nail down this issue. While I am at your email and going through the code to figure out the possible cause for it, unfortunately I don't see any script in the attachment of the bug. Could you please cross check? After I added some debug print,which like below, in glusterd-store.c and I found that the /var/lib/glusterd/vols/xxx/info and /var/lib/glusterd/vols/xxx/bricks/* are removed. But other files in /var/lib/glusterd/vols/xxx/ will not be remove. int32_t glusterd_store_volinfo
Re: [Gluster-users] question about info and info.tmp
On Fri, Nov 25, 2016 at 12:06 PM, songxinwrote: > Hi Atin, > Do you mean that you have the workaround applicable now? > Or it will take time to design the workaround? > > If you have workaround now, could you share it to me ? > If you end up in having a 0 byte info file you'd need to copy the same info file from other node and put it there and restart glusterd. > > Thanks, > Xin, > > > > > > 在 2016-11-24 19:12:07,"Atin Mukherjee" 写道: > > Xin - I appreciate your patience. I'd need some more time to pick this > item up from my backlog. I believe we have a workaround applicable here too. > > On Thu, 24 Nov 2016 at 14:24, songxin wrote: > >> >> >> >> Hi Atin, >> Actually, the glusterfs is used in my project. >> And our test team find this issue. >> So I want to make sure that whether you plan to fix it. >> if you have plan I will wait you because your method shoud be better than >> mine. >> >> Thanks, >> Xin >> >> >> 在 2016-11-21 10:00:36,"Atin Mukherjee" 写道: >> >> Hi Xin, >> >> I've not got a chance to look into it yet. delete stale volume function >> is in place to take care of wiping off volume configuration data which has >> been deleted from the cluster. However we need to revisit this code to see >> if this function is anymore needed given we recently added a validation to >> fail delete request if one of the glusterd is down. I'll get back to you on >> this. >> >> On Mon, 21 Nov 2016 at 07:24, songxin wrote: >> >> Hi Atin, >> Thank you for your support. >> >> And any conclusions about this issue? >> >> Thanks, >> Xin >> >> >> >> >> >> 在 2016-11-16 20:59:05,"Atin Mukherjee" 写道: >> >> >> >> On Tue, Nov 15, 2016 at 1:53 PM, songxin wrote: >> >> ok, thank you. >> >> >> >> >> 在 2016-11-15 16:12:34,"Atin Mukherjee" 写道: >> >> >> >> On Tue, Nov 15, 2016 at 12:47 PM, songxin wrote: >> >> >> Hi Atin, >> >> I think the root cause is in the function glusterd_import_friend_volume >> as below. >> >> int32_t >> glusterd_import_friend_volume (dict_t *peer_data, size_t count) >> { >> ... >> ret = glusterd_volinfo_find (new_volinfo->volname, _volinfo); >> if (0 == ret) { >> (void) gd_check_and_update_rebalance_info (old_volinfo, >>new_volinfo); >> (void) glusterd_delete_stale_volume (old_volinfo, >> new_volinfo); >> } >> ... >> ret = glusterd_store_volinfo (new_volinfo, >> GLUSTERD_VOLINFO_VER_AC_NONE); >> if (ret) { >> gf_msg (this->name, GF_LOG_ERROR, 0, >> GD_MSG_VOLINFO_STORE_FAIL, "Failed to store " >> "volinfo for volume %s", new_volinfo->volname); >> goto out; >> } >> ... >> } >> >> glusterd_delete_stale_volume will remove the info and bricks/* and the >> glusterd_store_volinfo will create the new one. >> But if glusterd is killed before rename the info will is empty. >> >> And glusterd will start failed because the infois empty in the next time >> you start the glusterd. >> >> Any idea, Atin? >> >> >> Give me some time, will check it out, but reading at this analysis looks >> very well possible if a volume is changed when the glusterd was done on >> node a and when the same comes up during peer handshake we update the >> volinfo and during that time glusterd goes down once again. I'll confirm it >> by tomorrow. >> >> >> I checked the code and it does look like you have got the right RCA for >> the issue which you simulated through those two scripts. However this can >> happen even when you try to create a fresh volume and while glusterd tries >> to write the content into the store and goes down before renaming the >> info.tmp file you get into the same situation. >> >> I'd really need to think through if this can be fixed. Suggestions are >> always appreciated. >> >> >> >> >> BTW, excellent work Xin! >> >> >> Thanks, >> Xin >> >> >> 在 2016-11-15 12:07:05,"Atin Mukherjee" 写道: >> >> >> >> On Tue, Nov 15, 2016 at 8:58 AM, songxin wrote: >> >> Hi Atin, >> I have some clues about this issue. >> I could reproduce this issue use the scrip that mentioned in >> https://bugzilla.redhat.com/show_bug.cgi?id=1308487 . >> >> >> I really appreciate your help in trying to nail down this issue. While I >> am at your email and going through the code to figure out the possible >> cause for it, unfortunately I don't see any script in the attachment of the >> bug. Could you please cross check? >> >> >> >> After I added some debug print,which like below, in glusterd-store.c and >> I found that the /var/lib/glusterd/vols/xxx/info and >> /var/lib/glusterd/vols/xxx/bricks/* are removed. >> But other files in /var/lib/glusterd/vols/xxx/ will not be remove. >> >>
Re: [Gluster-users] question about info and info.tmp
Hi Atin, Do you mean that you have the workaround applicable now? Or it will take time to design the workaround? If you have workaround now, could you share it to me ? Thanks, Xin, 在 2016-11-24 19:12:07,"Atin Mukherjee"写道: Xin - I appreciate your patience. I'd need some more time to pick this item up from my backlog. I believe we have a workaround applicable here too. On Thu, 24 Nov 2016 at 14:24, songxin wrote: Hi Atin, Actually, the glusterfs is used in my project. And our test team find this issue. So I want to make sure that whether you plan to fix it. if you have plan I will wait you because your method shoud be better than mine. Thanks, Xin 在 2016-11-21 10:00:36,"Atin Mukherjee" 写道: Hi Xin, I've not got a chance to look into it yet. delete stale volume function is in place to take care of wiping off volume configuration data which has been deleted from the cluster. However we need to revisit this code to see if this function is anymore needed given we recently added a validation to fail delete request if one of the glusterd is down. I'll get back to you on this. On Mon, 21 Nov 2016 at 07:24, songxin wrote: Hi Atin, Thank you for your support. And any conclusions about this issue? Thanks, Xin 在 2016-11-16 20:59:05,"Atin Mukherjee" 写道: On Tue, Nov 15, 2016 at 1:53 PM, songxin wrote: ok, thank you. 在 2016-11-15 16:12:34,"Atin Mukherjee" 写道: On Tue, Nov 15, 2016 at 12:47 PM, songxin wrote: Hi Atin, I think the root cause is in the function glusterd_import_friend_volume as below. int32_t glusterd_import_friend_volume (dict_t *peer_data, size_t count) { ... ret = glusterd_volinfo_find (new_volinfo->volname, _volinfo); if (0 == ret) { (void) gd_check_and_update_rebalance_info (old_volinfo, new_volinfo); (void) glusterd_delete_stale_volume (old_volinfo, new_volinfo); } ... ret = glusterd_store_volinfo (new_volinfo, GLUSTERD_VOLINFO_VER_AC_NONE); if (ret) { gf_msg (this->name, GF_LOG_ERROR, 0, GD_MSG_VOLINFO_STORE_FAIL, "Failed to store " "volinfo for volume %s", new_volinfo->volname); goto out; } ... } glusterd_delete_stale_volume will remove the info and bricks/* and the glusterd_store_volinfo will create the new one. But if glusterd is killed before rename the info will is empty. And glusterd will start failed because the infois empty in the next time you start the glusterd. Any idea, Atin? Give me some time, will check it out, but reading at this analysis looks very well possible if a volume is changed when the glusterd was done on node a and when the same comes up during peer handshake we update the volinfo and during that time glusterd goes down once again. I'll confirm it by tomorrow. I checked the code and it does look like you have got the right RCA for the issue which you simulated through those two scripts. However this can happen even when you try to create a fresh volume and while glusterd tries to write the content into the store and goes down before renaming the info.tmp file you get into the same situation. I'd really need to think through if this can be fixed. Suggestions are always appreciated. BTW, excellent work Xin! Thanks, Xin 在 2016-11-15 12:07:05,"Atin Mukherjee" 写道: On Tue, Nov 15, 2016 at 8:58 AM, songxin wrote: Hi Atin, I have some clues about this issue. I could reproduce this issue use the scrip that mentioned in https://bugzilla.redhat.com/show_bug.cgi?id=1308487 . I really appreciate your help in trying to nail down this issue. While I am at your email and going through the code to figure out the possible cause for it, unfortunately I don't see any script in the attachment of the bug. Could you please cross check? After I added some debug print,which like below, in glusterd-store.c and I found that the /var/lib/glusterd/vols/xxx/info and /var/lib/glusterd/vols/xxx/bricks/* are removed. But other files in /var/lib/glusterd/vols/xxx/ will not be remove. int32_t glusterd_store_volinfo (glusterd_volinfo_t *volinfo, glusterd_volinfo_ver_ac_t ac) { int32_t ret = -1; GF_ASSERT (volinfo) ret = access("/var/lib/glusterd/vols/gv0/info", F_OK); if(ret < 0) { gf_msg (THIS->name, GF_LOG_ERROR, 0, 0, "info is not exit(%d)", errno); } else { ret = stat("/var/lib/glusterd/vols/gv0/info", ); if(ret < 0) { gf_msg (THIS->name,
Re: [Gluster-users] question about info and info.tmp
Xin - I appreciate your patience. I'd need some more time to pick this item up from my backlog. I believe we have a workaround applicable here too. On Thu, 24 Nov 2016 at 14:24, songxinwrote: > > > > Hi Atin, > Actually, the glusterfs is used in my project. > And our test team find this issue. > So I want to make sure that whether you plan to fix it. > if you have plan I will wait you because your method shoud be better than > mine. > > Thanks, > Xin > > > 在 2016-11-21 10:00:36,"Atin Mukherjee" 写道: > > Hi Xin, > > I've not got a chance to look into it yet. delete stale volume function is > in place to take care of wiping off volume configuration data which has > been deleted from the cluster. However we need to revisit this code to see > if this function is anymore needed given we recently added a validation to > fail delete request if one of the glusterd is down. I'll get back to you on > this. > > On Mon, 21 Nov 2016 at 07:24, songxin wrote: > > Hi Atin, > Thank you for your support. > > And any conclusions about this issue? > > Thanks, > Xin > > > > > > 在 2016-11-16 20:59:05,"Atin Mukherjee" 写道: > > > > On Tue, Nov 15, 2016 at 1:53 PM, songxin wrote: > > ok, thank you. > > > > > 在 2016-11-15 16:12:34,"Atin Mukherjee" 写道: > > > > On Tue, Nov 15, 2016 at 12:47 PM, songxin wrote: > > > Hi Atin, > > I think the root cause is in the function glusterd_import_friend_volume as > below. > > int32_t > glusterd_import_friend_volume (dict_t *peer_data, size_t count) > { > ... > ret = glusterd_volinfo_find (new_volinfo->volname, _volinfo); > if (0 == ret) { > (void) gd_check_and_update_rebalance_info (old_volinfo, >new_volinfo); > (void) glusterd_delete_stale_volume (old_volinfo, > new_volinfo); > } > ... > ret = glusterd_store_volinfo (new_volinfo, > GLUSTERD_VOLINFO_VER_AC_NONE); > if (ret) { > gf_msg (this->name, GF_LOG_ERROR, 0, > GD_MSG_VOLINFO_STORE_FAIL, "Failed to store " > "volinfo for volume %s", new_volinfo->volname); > goto out; > } > ... > } > > glusterd_delete_stale_volume will remove the info and bricks/* and the > glusterd_store_volinfo will create the new one. > But if glusterd is killed before rename the info will is empty. > > And glusterd will start failed because the infois empty in the next time > you start the glusterd. > > Any idea, Atin? > > > Give me some time, will check it out, but reading at this analysis looks > very well possible if a volume is changed when the glusterd was done on > node a and when the same comes up during peer handshake we update the > volinfo and during that time glusterd goes down once again. I'll confirm it > by tomorrow. > > > I checked the code and it does look like you have got the right RCA for > the issue which you simulated through those two scripts. However this can > happen even when you try to create a fresh volume and while glusterd tries > to write the content into the store and goes down before renaming the > info.tmp file you get into the same situation. > > I'd really need to think through if this can be fixed. Suggestions are > always appreciated. > > > > > BTW, excellent work Xin! > > > Thanks, > Xin > > > 在 2016-11-15 12:07:05,"Atin Mukherjee" 写道: > > > > On Tue, Nov 15, 2016 at 8:58 AM, songxin wrote: > > Hi Atin, > I have some clues about this issue. > I could reproduce this issue use the scrip that mentioned in > https://bugzilla.redhat.com/show_bug.cgi?id=1308487 . > > > I really appreciate your help in trying to nail down this issue. While I > am at your email and going through the code to figure out the possible > cause for it, unfortunately I don't see any script in the attachment of the > bug. Could you please cross check? > > > > After I added some debug print,which like below, in glusterd-store.c and I > found that the /var/lib/glusterd/vols/xxx/info and > /var/lib/glusterd/vols/xxx/bricks/* > are removed. > But other files in /var/lib/glusterd/vols/xxx/ will not be remove. > > int32_t > glusterd_store_volinfo (glusterd_volinfo_t *volinfo, > glusterd_volinfo_ver_ac_t ac) > { > int32_t ret = -1; > > GF_ASSERT (volinfo) > > ret = access("/var/lib/glusterd/vols/gv0/info", F_OK); > if(ret < 0) > { > gf_msg (THIS->name, GF_LOG_ERROR, 0, 0, "info is not > exit(%d)", errno); > } > else > { > ret = stat("/var/lib/glusterd/vols/gv0/info", ); > if(ret < 0) > { > gf_msg (THIS->name, GF_LOG_ERROR, 0, 0, "stat info > error"); > } >
Re: [Gluster-users] question about info and info.tmp
Hi Atin, Actually, the glusterfs is used in my project. And our test team find this issue. So I want to make sure that whether you plan to fix it. if you have plan I will wait you because your method shoud be better than mine. Thanks, Xin 在 2016-11-21 10:00:36,"Atin Mukherjee"写道: Hi Xin, I've not got a chance to look into it yet. delete stale volume function is in place to take care of wiping off volume configuration data which has been deleted from the cluster. However we need to revisit this code to see if this function is anymore needed given we recently added a validation to fail delete request if one of the glusterd is down. I'll get back to you on this. On Mon, 21 Nov 2016 at 07:24, songxin wrote: Hi Atin, Thank you for your support. And any conclusions about this issue? Thanks, Xin 在 2016-11-16 20:59:05,"Atin Mukherjee" 写道: On Tue, Nov 15, 2016 at 1:53 PM, songxin wrote: ok, thank you. 在 2016-11-15 16:12:34,"Atin Mukherjee" 写道: On Tue, Nov 15, 2016 at 12:47 PM, songxin wrote: Hi Atin, I think the root cause is in the function glusterd_import_friend_volume as below. int32_t glusterd_import_friend_volume (dict_t *peer_data, size_t count) { ... ret = glusterd_volinfo_find (new_volinfo->volname, _volinfo); if (0 == ret) { (void) gd_check_and_update_rebalance_info (old_volinfo, new_volinfo); (void) glusterd_delete_stale_volume (old_volinfo, new_volinfo); } ... ret = glusterd_store_volinfo (new_volinfo, GLUSTERD_VOLINFO_VER_AC_NONE); if (ret) { gf_msg (this->name, GF_LOG_ERROR, 0, GD_MSG_VOLINFO_STORE_FAIL, "Failed to store " "volinfo for volume %s", new_volinfo->volname); goto out; } ... } glusterd_delete_stale_volume will remove the info and bricks/* and the glusterd_store_volinfo will create the new one. But if glusterd is killed before rename the info will is empty. And glusterd will start failed because the infois empty in the next time you start the glusterd. Any idea, Atin? Give me some time, will check it out, but reading at this analysis looks very well possible if a volume is changed when the glusterd was done on node a and when the same comes up during peer handshake we update the volinfo and during that time glusterd goes down once again. I'll confirm it by tomorrow. I checked the code and it does look like you have got the right RCA for the issue which you simulated through those two scripts. However this can happen even when you try to create a fresh volume and while glusterd tries to write the content into the store and goes down before renaming the info.tmp file you get into the same situation. I'd really need to think through if this can be fixed. Suggestions are always appreciated. BTW, excellent work Xin! Thanks, Xin 在 2016-11-15 12:07:05,"Atin Mukherjee" 写道: On Tue, Nov 15, 2016 at 8:58 AM, songxin wrote: Hi Atin, I have some clues about this issue. I could reproduce this issue use the scrip that mentioned in https://bugzilla.redhat.com/show_bug.cgi?id=1308487 . I really appreciate your help in trying to nail down this issue. While I am at your email and going through the code to figure out the possible cause for it, unfortunately I don't see any script in the attachment of the bug. Could you please cross check? After I added some debug print,which like below, in glusterd-store.c and I found that the /var/lib/glusterd/vols/xxx/info and /var/lib/glusterd/vols/xxx/bricks/* are removed. But other files in /var/lib/glusterd/vols/xxx/ will not be remove. int32_t glusterd_store_volinfo (glusterd_volinfo_t *volinfo, glusterd_volinfo_ver_ac_t ac) { int32_t ret = -1; GF_ASSERT (volinfo) ret = access("/var/lib/glusterd/vols/gv0/info", F_OK); if(ret < 0) { gf_msg (THIS->name, GF_LOG_ERROR, 0, 0, "info is not exit(%d)", errno); } else { ret = stat("/var/lib/glusterd/vols/gv0/info", ); if(ret < 0) { gf_msg (THIS->name, GF_LOG_ERROR, 0, 0, "stat info error"); } else { gf_msg (THIS->name, GF_LOG_ERROR, 0, 0, "info size is %lu, inode num is %lu", buf.st_size, buf.st_ino); } } glusterd_perform_volinfo_version_action (volinfo, ac); ret = glusterd_store_create_volume_dir (volinfo); if (ret) goto out; ... } So it is easy to understand why the info or
Re: [Gluster-users] question about info and info.tmp
Hi Atin, Ok.Thank you for your reply. Thanks, Xin 在 2016-11-21 10:00:36,"Atin Mukherjee"写道: Hi Xin, I've not got a chance to look into it yet. delete stale volume function is in place to take care of wiping off volume configuration data which has been deleted from the cluster. However we need to revisit this code to see if this function is anymore needed given we recently added a validation to fail delete request if one of the glusterd is down. I'll get back to you on this. On Mon, 21 Nov 2016 at 07:24, songxin wrote: Hi Atin, Thank you for your support. And any conclusions about this issue? Thanks, Xin 在 2016-11-16 20:59:05,"Atin Mukherjee" 写道: On Tue, Nov 15, 2016 at 1:53 PM, songxin wrote: ok, thank you. 在 2016-11-15 16:12:34,"Atin Mukherjee" 写道: On Tue, Nov 15, 2016 at 12:47 PM, songxin wrote: Hi Atin, I think the root cause is in the function glusterd_import_friend_volume as below. int32_t glusterd_import_friend_volume (dict_t *peer_data, size_t count) { ... ret = glusterd_volinfo_find (new_volinfo->volname, _volinfo); if (0 == ret) { (void) gd_check_and_update_rebalance_info (old_volinfo, new_volinfo); (void) glusterd_delete_stale_volume (old_volinfo, new_volinfo); } ... ret = glusterd_store_volinfo (new_volinfo, GLUSTERD_VOLINFO_VER_AC_NONE); if (ret) { gf_msg (this->name, GF_LOG_ERROR, 0, GD_MSG_VOLINFO_STORE_FAIL, "Failed to store " "volinfo for volume %s", new_volinfo->volname); goto out; } ... } glusterd_delete_stale_volume will remove the info and bricks/* and the glusterd_store_volinfo will create the new one. But if glusterd is killed before rename the info will is empty. And glusterd will start failed because the infois empty in the next time you start the glusterd. Any idea, Atin? Give me some time, will check it out, but reading at this analysis looks very well possible if a volume is changed when the glusterd was done on node a and when the same comes up during peer handshake we update the volinfo and during that time glusterd goes down once again. I'll confirm it by tomorrow. I checked the code and it does look like you have got the right RCA for the issue which you simulated through those two scripts. However this can happen even when you try to create a fresh volume and while glusterd tries to write the content into the store and goes down before renaming the info.tmp file you get into the same situation. I'd really need to think through if this can be fixed. Suggestions are always appreciated. BTW, excellent work Xin! Thanks, Xin 在 2016-11-15 12:07:05,"Atin Mukherjee" 写道: On Tue, Nov 15, 2016 at 8:58 AM, songxin wrote: Hi Atin, I have some clues about this issue. I could reproduce this issue use the scrip that mentioned in https://bugzilla.redhat.com/show_bug.cgi?id=1308487 . I really appreciate your help in trying to nail down this issue. While I am at your email and going through the code to figure out the possible cause for it, unfortunately I don't see any script in the attachment of the bug. Could you please cross check? After I added some debug print,which like below, in glusterd-store.c and I found that the /var/lib/glusterd/vols/xxx/info and /var/lib/glusterd/vols/xxx/bricks/* are removed. But other files in /var/lib/glusterd/vols/xxx/ will not be remove. int32_t glusterd_store_volinfo (glusterd_volinfo_t *volinfo, glusterd_volinfo_ver_ac_t ac) { int32_t ret = -1; GF_ASSERT (volinfo) ret = access("/var/lib/glusterd/vols/gv0/info", F_OK); if(ret < 0) { gf_msg (THIS->name, GF_LOG_ERROR, 0, 0, "info is not exit(%d)", errno); } else { ret = stat("/var/lib/glusterd/vols/gv0/info", ); if(ret < 0) { gf_msg (THIS->name, GF_LOG_ERROR, 0, 0, "stat info error"); } else { gf_msg (THIS->name, GF_LOG_ERROR, 0, 0, "info size is %lu, inode num is %lu", buf.st_size, buf.st_ino); } } glusterd_perform_volinfo_version_action (volinfo, ac); ret = glusterd_store_create_volume_dir (volinfo); if (ret) goto out; ... } So it is easy to understand why the info or 10.32.1.144.-opt-lvmdir-c2-brick sometimes is empty. It is becaue the info file is not exist, and it will be create by “fd = open (path, O_RDWR | O_CREAT | O_APPEND, 0600);” in function
Re: [Gluster-users] question about info and info.tmp
Hi Xin, I've not got a chance to look into it yet. delete stale volume function is in place to take care of wiping off volume configuration data which has been deleted from the cluster. However we need to revisit this code to see if this function is anymore needed given we recently added a validation to fail delete request if one of the glusterd is down. I'll get back to you on this. On Mon, 21 Nov 2016 at 07:24, songxinwrote: > Hi Atin, > Thank you for your support. > > And any conclusions about this issue? > > Thanks, > Xin > > > > > > 在 2016-11-16 20:59:05,"Atin Mukherjee" 写道: > > > > On Tue, Nov 15, 2016 at 1:53 PM, songxin wrote: > > ok, thank you. > > > > > 在 2016-11-15 16:12:34,"Atin Mukherjee" 写道: > > > > On Tue, Nov 15, 2016 at 12:47 PM, songxin wrote: > > > Hi Atin, > > I think the root cause is in the function glusterd_import_friend_volume as > below. > > int32_t > glusterd_import_friend_volume (dict_t *peer_data, size_t count) > { > ... > ret = glusterd_volinfo_find (new_volinfo->volname, _volinfo); > if (0 == ret) { > (void) gd_check_and_update_rebalance_info (old_volinfo, >new_volinfo); > (void) glusterd_delete_stale_volume (old_volinfo, > new_volinfo); > } > ... > ret = glusterd_store_volinfo (new_volinfo, > GLUSTERD_VOLINFO_VER_AC_NONE); > if (ret) { > gf_msg (this->name, GF_LOG_ERROR, 0, > GD_MSG_VOLINFO_STORE_FAIL, "Failed to store " > "volinfo for volume %s", new_volinfo->volname); > goto out; > } > ... > } > > glusterd_delete_stale_volume will remove the info and bricks/* and the > glusterd_store_volinfo will create the new one. > But if glusterd is killed before rename the info will is empty. > > And glusterd will start failed because the infois empty in the next time > you start the glusterd. > > Any idea, Atin? > > > Give me some time, will check it out, but reading at this analysis looks > very well possible if a volume is changed when the glusterd was done on > node a and when the same comes up during peer handshake we update the > volinfo and during that time glusterd goes down once again. I'll confirm it > by tomorrow. > > > I checked the code and it does look like you have got the right RCA for > the issue which you simulated through those two scripts. However this can > happen even when you try to create a fresh volume and while glusterd tries > to write the content into the store and goes down before renaming the > info.tmp file you get into the same situation. > > I'd really need to think through if this can be fixed. Suggestions are > always appreciated. > > > > > BTW, excellent work Xin! > > > Thanks, > Xin > > > 在 2016-11-15 12:07:05,"Atin Mukherjee" 写道: > > > > On Tue, Nov 15, 2016 at 8:58 AM, songxin wrote: > > Hi Atin, > I have some clues about this issue. > I could reproduce this issue use the scrip that mentioned in > https://bugzilla.redhat.com/show_bug.cgi?id=1308487 . > > > I really appreciate your help in trying to nail down this issue. While I > am at your email and going through the code to figure out the possible > cause for it, unfortunately I don't see any script in the attachment of the > bug. Could you please cross check? > > > > After I added some debug print,which like below, in glusterd-store.c and I > found that the /var/lib/glusterd/vols/xxx/info and > /var/lib/glusterd/vols/xxx/bricks/* > are removed. > But other files in /var/lib/glusterd/vols/xxx/ will not be remove. > > int32_t > glusterd_store_volinfo (glusterd_volinfo_t *volinfo, > glusterd_volinfo_ver_ac_t ac) > { > int32_t ret = -1; > > GF_ASSERT (volinfo) > > ret = access("/var/lib/glusterd/vols/gv0/info", F_OK); > if(ret < 0) > { > gf_msg (THIS->name, GF_LOG_ERROR, 0, 0, "info is not > exit(%d)", errno); > } > else > { > ret = stat("/var/lib/glusterd/vols/gv0/info", ); > if(ret < 0) > { > gf_msg (THIS->name, GF_LOG_ERROR, 0, 0, "stat info > error"); > } > else > { > gf_msg (THIS->name, GF_LOG_ERROR, 0, 0, "info size > is %lu, inode num is %lu", buf.st_size, buf.st_ino); > } > } > > glusterd_perform_volinfo_version_action (volinfo, ac); > ret = glusterd_store_create_volume_dir (volinfo); > if (ret) > goto out; > > ... > } > > So it is easy to understand why the info or > 10.32.1.144.-opt-lvmdir-c2-brick sometimes is empty. > It is becaue the info file is not exist, and it will be create by “fd = > open (path, O_RDWR
Re: [Gluster-users] question about info and info.tmp
Hi Atin, Thank you for your support. And any conclusions about this issue? Thanks, Xin 在 2016-11-16 20:59:05,"Atin Mukherjee"写道: On Tue, Nov 15, 2016 at 1:53 PM, songxin wrote: ok, thank you. 在 2016-11-15 16:12:34,"Atin Mukherjee" 写道: On Tue, Nov 15, 2016 at 12:47 PM, songxin wrote: Hi Atin, I think the root cause is in the function glusterd_import_friend_volume as below. int32_t glusterd_import_friend_volume (dict_t *peer_data, size_t count) { ... ret = glusterd_volinfo_find (new_volinfo->volname, _volinfo); if (0 == ret) { (void) gd_check_and_update_rebalance_info (old_volinfo, new_volinfo); (void) glusterd_delete_stale_volume (old_volinfo, new_volinfo); } ... ret = glusterd_store_volinfo (new_volinfo, GLUSTERD_VOLINFO_VER_AC_NONE); if (ret) { gf_msg (this->name, GF_LOG_ERROR, 0, GD_MSG_VOLINFO_STORE_FAIL, "Failed to store " "volinfo for volume %s", new_volinfo->volname); goto out; } ... } glusterd_delete_stale_volume will remove the info and bricks/* and the glusterd_store_volinfo will create the new one. But if glusterd is killed before rename the info will is empty. And glusterd will start failed because the infois empty in the next time you start the glusterd. Any idea, Atin? Give me some time, will check it out, but reading at this analysis looks very well possible if a volume is changed when the glusterd was done on node a and when the same comes up during peer handshake we update the volinfo and during that time glusterd goes down once again. I'll confirm it by tomorrow. I checked the code and it does look like you have got the right RCA for the issue which you simulated through those two scripts. However this can happen even when you try to create a fresh volume and while glusterd tries to write the content into the store and goes down before renaming the info.tmp file you get into the same situation. I'd really need to think through if this can be fixed. Suggestions are always appreciated. BTW, excellent work Xin! Thanks, Xin 在 2016-11-15 12:07:05,"Atin Mukherjee" 写道: On Tue, Nov 15, 2016 at 8:58 AM, songxin wrote: Hi Atin, I have some clues about this issue. I could reproduce this issue use the scrip that mentioned in https://bugzilla.redhat.com/show_bug.cgi?id=1308487 . I really appreciate your help in trying to nail down this issue. While I am at your email and going through the code to figure out the possible cause for it, unfortunately I don't see any script in the attachment of the bug. Could you please cross check? After I added some debug print,which like below, in glusterd-store.c and I found that the /var/lib/glusterd/vols/xxx/info and /var/lib/glusterd/vols/xxx/bricks/* are removed. But other files in /var/lib/glusterd/vols/xxx/ will not be remove. int32_t glusterd_store_volinfo (glusterd_volinfo_t *volinfo, glusterd_volinfo_ver_ac_t ac) { int32_t ret = -1; GF_ASSERT (volinfo) ret = access("/var/lib/glusterd/vols/gv0/info", F_OK); if(ret < 0) { gf_msg (THIS->name, GF_LOG_ERROR, 0, 0, "info is not exit(%d)", errno); } else { ret = stat("/var/lib/glusterd/vols/gv0/info", ); if(ret < 0) { gf_msg (THIS->name, GF_LOG_ERROR, 0, 0, "stat info error"); } else { gf_msg (THIS->name, GF_LOG_ERROR, 0, 0, "info size is %lu, inode num is %lu", buf.st_size, buf.st_ino); } } glusterd_perform_volinfo_version_action (volinfo, ac); ret = glusterd_store_create_volume_dir (volinfo); if (ret) goto out; ... } So it is easy to understand why the info or 10.32.1.144.-opt-lvmdir-c2-brick sometimes is empty. It is becaue the info file is not exist, and it will be create by “fd = open (path, O_RDWR | O_CREAT | O_APPEND, 0600);” in function gf_store_handle_new. And the info file is empty before rename. So the info file is empty if glusterd shutdown before rename. My question is following. 1.I did not find the point the info is removed.Could you tell me the point where the info and /bricks/* are removed? 2.why the file info and bricks/* is removed?But other files in var/lib/glusterd/vols/xxx/ are not be removed? AFAIK, we never delete the info file and hence this file is opened with O_APPEND flag. As I said I will go back and cross check the code once again. Thanks, Xin 在 2016-11-11 20:34:05,"Atin Mukherjee"
Re: [Gluster-users] question about info and info.tmp
Hi Atin, Thank you for your support. I have a question for you. glusterd_store_volinfo() will hidden remove the info and bricks/* by rename(). Why glusterd must remove the info and bricks/* in function glusterd_delete_stale_volume() before calling glusterd_store_volinfo()? Thanks, Xin 在 2016-11-16 20:59:05,"Atin Mukherjee"写道: On Tue, Nov 15, 2016 at 1:53 PM, songxin wrote: ok, thank you. 在 2016-11-15 16:12:34,"Atin Mukherjee" 写道: On Tue, Nov 15, 2016 at 12:47 PM, songxin wrote: Hi Atin, I think the root cause is in the function glusterd_import_friend_volume as below. int32_t glusterd_import_friend_volume (dict_t *peer_data, size_t count) { ... ret = glusterd_volinfo_find (new_volinfo->volname, _volinfo); if (0 == ret) { (void) gd_check_and_update_rebalance_info (old_volinfo, new_volinfo); (void) glusterd_delete_stale_volume (old_volinfo, new_volinfo); } ... ret = glusterd_store_volinfo (new_volinfo, GLUSTERD_VOLINFO_VER_AC_NONE); if (ret) { gf_msg (this->name, GF_LOG_ERROR, 0, GD_MSG_VOLINFO_STORE_FAIL, "Failed to store " "volinfo for volume %s", new_volinfo->volname); goto out; } ... } glusterd_delete_stale_volume will remove the info and bricks/* and the glusterd_store_volinfo will create the new one. But if glusterd is killed before rename the info will is empty. And glusterd will start failed because the infois empty in the next time you start the glusterd. Any idea, Atin? Give me some time, will check it out, but reading at this analysis looks very well possible if a volume is changed when the glusterd was done on node a and when the same comes up during peer handshake we update the volinfo and during that time glusterd goes down once again. I'll confirm it by tomorrow. I checked the code and it does look like you have got the right RCA for the issue which you simulated through those two scripts. However this can happen even when you try to create a fresh volume and while glusterd tries to write the content into the store and goes down before renaming the info.tmp file you get into the same situation. I'd really need to think through if this can be fixed. Suggestions are always appreciated. BTW, excellent work Xin! Thanks, Xin 在 2016-11-15 12:07:05,"Atin Mukherjee" 写道: On Tue, Nov 15, 2016 at 8:58 AM, songxin wrote: Hi Atin, I have some clues about this issue. I could reproduce this issue use the scrip that mentioned in https://bugzilla.redhat.com/show_bug.cgi?id=1308487 . I really appreciate your help in trying to nail down this issue. While I am at your email and going through the code to figure out the possible cause for it, unfortunately I don't see any script in the attachment of the bug. Could you please cross check? After I added some debug print,which like below, in glusterd-store.c and I found that the /var/lib/glusterd/vols/xxx/info and /var/lib/glusterd/vols/xxx/bricks/* are removed. But other files in /var/lib/glusterd/vols/xxx/ will not be remove. int32_t glusterd_store_volinfo (glusterd_volinfo_t *volinfo, glusterd_volinfo_ver_ac_t ac) { int32_t ret = -1; GF_ASSERT (volinfo) ret = access("/var/lib/glusterd/vols/gv0/info", F_OK); if(ret < 0) { gf_msg (THIS->name, GF_LOG_ERROR, 0, 0, "info is not exit(%d)", errno); } else { ret = stat("/var/lib/glusterd/vols/gv0/info", ); if(ret < 0) { gf_msg (THIS->name, GF_LOG_ERROR, 0, 0, "stat info error"); } else { gf_msg (THIS->name, GF_LOG_ERROR, 0, 0, "info size is %lu, inode num is %lu", buf.st_size, buf.st_ino); } } glusterd_perform_volinfo_version_action (volinfo, ac); ret = glusterd_store_create_volume_dir (volinfo); if (ret) goto out; ... } So it is easy to understand why the info or 10.32.1.144.-opt-lvmdir-c2-brick sometimes is empty. It is becaue the info file is not exist, and it will be create by “fd = open (path, O_RDWR | O_CREAT | O_APPEND, 0600);” in function gf_store_handle_new. And the info file is empty before rename. So the info file is empty if glusterd shutdown before rename. My question is following. 1.I did not find the point the info is removed.Could you tell me the point where the info and /bricks/* are removed? 2.why the file info and bricks/* is removed?But other files in var/lib/glusterd/vols/xxx/ are not be removed? AFAIK, we never
Re: [Gluster-users] question about info and info.tmp
On Tue, Nov 15, 2016 at 1:53 PM, songxinwrote: > ok, thank you. > > > > > 在 2016-11-15 16:12:34,"Atin Mukherjee" 写道: > > > > On Tue, Nov 15, 2016 at 12:47 PM, songxin wrote: > >> >> Hi Atin, >> >> I think the root cause is in the function glusterd_import_friend_volume >> as below. >> >> int32_t >> glusterd_import_friend_volume (dict_t *peer_data, size_t count) >> { >> ... >> ret = glusterd_volinfo_find (new_volinfo->volname, _volinfo); >> if (0 == ret) { >> (void) gd_check_and_update_rebalance_info (old_volinfo, >>new_volinfo); >> (void) glusterd_delete_stale_volume (old_volinfo, >> new_volinfo); >> } >> ... >> ret = glusterd_store_volinfo (new_volinfo, >> GLUSTERD_VOLINFO_VER_AC_NONE); >> if (ret) { >> gf_msg (this->name, GF_LOG_ERROR, 0, >> GD_MSG_VOLINFO_STORE_FAIL, "Failed to store " >> "volinfo for volume %s", new_volinfo->volname); >> goto out; >> } >> ... >> } >> >> glusterd_delete_stale_volume will remove the info and bricks/* and the >> glusterd_store_volinfo will create the new one. >> But if glusterd is killed before rename the info will is empty. >> >> And glusterd will start failed because the infois empty in the next time >> you start the glusterd. >> >> Any idea, Atin? >> > > Give me some time, will check it out, but reading at this analysis looks > very well possible if a volume is changed when the glusterd was done on > node a and when the same comes up during peer handshake we update the > volinfo and during that time glusterd goes down once again. I'll confirm it > by tomorrow. > > I checked the code and it does look like you have got the right RCA for the issue which you simulated through those two scripts. However this can happen even when you try to create a fresh volume and while glusterd tries to write the content into the store and goes down before renaming the info.tmp file you get into the same situation. I'd really need to think through if this can be fixed. Suggestions are always appreciated. > > BTW, excellent work Xin! > > >> Thanks, >> Xin >> >> >> 在 2016-11-15 12:07:05,"Atin Mukherjee" 写道: >> >> >> >> On Tue, Nov 15, 2016 at 8:58 AM, songxin wrote: >> >>> Hi Atin, >>> I have some clues about this issue. >>> I could reproduce this issue use the scrip that mentioned in >>> https://bugzilla.redhat.com/show_bug.cgi?id=1308487 . >>> >> >> I really appreciate your help in trying to nail down this issue. While I >> am at your email and going through the code to figure out the possible >> cause for it, unfortunately I don't see any script in the attachment of the >> bug. Could you please cross check? >> >> >>> >>> After I added some debug print,which like below, in glusterd-store.c and >>> I found that the /var/lib/glusterd/vols/xxx/info and >>> /var/lib/glusterd/vols/xxx/bricks/* are removed. >>> But other files in /var/lib/glusterd/vols/xxx/ will not be remove. >>> >>> int32_t >>> glusterd_store_volinfo (glusterd_volinfo_t *volinfo, >>> glusterd_volinfo_ver_ac_t ac) >>> { >>> int32_t ret = -1; >>> >>> GF_ASSERT (volinfo) >>> >>> ret = access("/var/lib/glusterd/vols/gv0/info", F_OK); >>> if(ret < 0) >>> { >>> gf_msg (THIS->name, GF_LOG_ERROR, 0, 0, "info is not >>> exit(%d)", errno); >>> } >>> else >>> { >>> ret = stat("/var/lib/glusterd/vols/gv0/info", ); >>> if(ret < 0) >>> { >>> gf_msg (THIS->name, GF_LOG_ERROR, 0, 0, "stat >>> info error"); >>> } >>> else >>> { >>> gf_msg (THIS->name, GF_LOG_ERROR, 0, 0, "info >>> size is %lu, inode num is %lu", buf.st_size, buf.st_ino); >>> } >>> } >>> >>> glusterd_perform_volinfo_version_action (volinfo, ac); >>> ret = glusterd_store_create_volume_dir (volinfo); >>> if (ret) >>> goto out; >>> >>> ... >>> } >>> >>> So it is easy to understand why the info or 10.32.1.144.-opt-lvmdir-c2- >>> brick sometimes is empty. >>> It is becaue the info file is not exist, and it will be create by “fd = >>> open (path, O_RDWR | O_CREAT | O_APPEND, 0600);” in function >>> gf_store_handle_new. >>> And the info file is empty before rename. >>> So the info file is empty if glusterd shutdown before rename. >>> >>> >> >>> My question is following. >>> 1.I did not find the point the info is removed.Could you tell me the >>> point where the info and /bricks/* are removed? >>> 2.why the file info and bricks/* is removed?But other files in >>> var/lib/glusterd/vols/xxx/ >>> are not be removed? >>> >> >> AFAIK, we never delete the
Re: [Gluster-users] question about info and info.tmp
ok, thank you. 在 2016-11-15 16:12:34,"Atin Mukherjee"写道: On Tue, Nov 15, 2016 at 12:47 PM, songxin wrote: Hi Atin, I think the root cause is in the function glusterd_import_friend_volume as below. int32_t glusterd_import_friend_volume (dict_t *peer_data, size_t count) { ... ret = glusterd_volinfo_find (new_volinfo->volname, _volinfo); if (0 == ret) { (void) gd_check_and_update_rebalance_info (old_volinfo, new_volinfo); (void) glusterd_delete_stale_volume (old_volinfo, new_volinfo); } ... ret = glusterd_store_volinfo (new_volinfo, GLUSTERD_VOLINFO_VER_AC_NONE); if (ret) { gf_msg (this->name, GF_LOG_ERROR, 0, GD_MSG_VOLINFO_STORE_FAIL, "Failed to store " "volinfo for volume %s", new_volinfo->volname); goto out; } ... } glusterd_delete_stale_volume will remove the info and bricks/* and the glusterd_store_volinfo will create the new one. But if glusterd is killed before rename the info will is empty. And glusterd will start failed because the infois empty in the next time you start the glusterd. Any idea, Atin? Give me some time, will check it out, but reading at this analysis looks very well possible if a volume is changed when the glusterd was done on node a and when the same comes up during peer handshake we update the volinfo and during that time glusterd goes down once again. I'll confirm it by tomorrow. BTW, excellent work Xin! Thanks, Xin 在 2016-11-15 12:07:05,"Atin Mukherjee" 写道: On Tue, Nov 15, 2016 at 8:58 AM, songxin wrote: Hi Atin, I have some clues about this issue. I could reproduce this issue use the scrip that mentioned in https://bugzilla.redhat.com/show_bug.cgi?id=1308487 . I really appreciate your help in trying to nail down this issue. While I am at your email and going through the code to figure out the possible cause for it, unfortunately I don't see any script in the attachment of the bug. Could you please cross check? After I added some debug print,which like below, in glusterd-store.c and I found that the /var/lib/glusterd/vols/xxx/info and /var/lib/glusterd/vols/xxx/bricks/* are removed. But other files in /var/lib/glusterd/vols/xxx/ will not be remove. int32_t glusterd_store_volinfo (glusterd_volinfo_t *volinfo, glusterd_volinfo_ver_ac_t ac) { int32_t ret = -1; GF_ASSERT (volinfo) ret = access("/var/lib/glusterd/vols/gv0/info", F_OK); if(ret < 0) { gf_msg (THIS->name, GF_LOG_ERROR, 0, 0, "info is not exit(%d)", errno); } else { ret = stat("/var/lib/glusterd/vols/gv0/info", ); if(ret < 0) { gf_msg (THIS->name, GF_LOG_ERROR, 0, 0, "stat info error"); } else { gf_msg (THIS->name, GF_LOG_ERROR, 0, 0, "info size is %lu, inode num is %lu", buf.st_size, buf.st_ino); } } glusterd_perform_volinfo_version_action (volinfo, ac); ret = glusterd_store_create_volume_dir (volinfo); if (ret) goto out; ... } So it is easy to understand why the info or 10.32.1.144.-opt-lvmdir-c2-brick sometimes is empty. It is becaue the info file is not exist, and it will be create by “fd = open (path, O_RDWR | O_CREAT | O_APPEND, 0600);” in function gf_store_handle_new. And the info file is empty before rename. So the info file is empty if glusterd shutdown before rename. My question is following. 1.I did not find the point the info is removed.Could you tell me the point where the info and /bricks/* are removed? 2.why the file info and bricks/* is removed?But other files in var/lib/glusterd/vols/xxx/ are not be removed? AFAIK, we never delete the info file and hence this file is opened with O_APPEND flag. As I said I will go back and cross check the code once again. Thanks, Xin 在 2016-11-11 20:34:05,"Atin Mukherjee" 写道: On Fri, Nov 11, 2016 at 4:00 PM, songxin wrote: Hi Atin, Thank you for your support. Sincerely wait for your reply. By the way, could you make sure that the issue, file info is empty, cause by rename is interrupted in kernel? As per my RCA on that bug, it looked to be. Thanks, Xin 在 2016-11-11 15:49:02,"Atin Mukherjee" 写道: On Fri, Nov 11, 2016 at 1:15 PM, songxin wrote: Hi Atin, Thank you for your reply. Actually it is very difficult to reproduce because I don't know when there was an ongoing commit happening.It is just a coincidence. But I want to make sure the root
Re: [Gluster-users] question about info and info.tmp
On Tue, Nov 15, 2016 at 12:47 PM, songxinwrote: > > Hi Atin, > > I think the root cause is in the function glusterd_import_friend_volume as > below. > > int32_t > glusterd_import_friend_volume (dict_t *peer_data, size_t count) > { > ... > ret = glusterd_volinfo_find (new_volinfo->volname, _volinfo); > if (0 == ret) { > (void) gd_check_and_update_rebalance_info (old_volinfo, >new_volinfo); > (void) glusterd_delete_stale_volume (old_volinfo, > new_volinfo); > } > ... > ret = glusterd_store_volinfo (new_volinfo, > GLUSTERD_VOLINFO_VER_AC_NONE); > if (ret) { > gf_msg (this->name, GF_LOG_ERROR, 0, > GD_MSG_VOLINFO_STORE_FAIL, "Failed to store " > "volinfo for volume %s", new_volinfo->volname); > goto out; > } > ... > } > > glusterd_delete_stale_volume will remove the info and bricks/* and the > glusterd_store_volinfo will create the new one. > But if glusterd is killed before rename the info will is empty. > > And glusterd will start failed because the infois empty in the next time > you start the glusterd. > > Any idea, Atin? > Give me some time, will check it out, but reading at this analysis looks very well possible if a volume is changed when the glusterd was done on node a and when the same comes up during peer handshake we update the volinfo and during that time glusterd goes down once again. I'll confirm it by tomorrow. BTW, excellent work Xin! > Thanks, > Xin > > > 在 2016-11-15 12:07:05,"Atin Mukherjee" 写道: > > > > On Tue, Nov 15, 2016 at 8:58 AM, songxin wrote: > >> Hi Atin, >> I have some clues about this issue. >> I could reproduce this issue use the scrip that mentioned in >> https://bugzilla.redhat.com/show_bug.cgi?id=1308487 . >> > > I really appreciate your help in trying to nail down this issue. While I > am at your email and going through the code to figure out the possible > cause for it, unfortunately I don't see any script in the attachment of the > bug. Could you please cross check? > > >> >> After I added some debug print,which like below, in glusterd-store.c and >> I found that the /var/lib/glusterd/vols/xxx/info and >> /var/lib/glusterd/vols/xxx/bricks/* are removed. >> But other files in /var/lib/glusterd/vols/xxx/ will not be remove. >> >> int32_t >> glusterd_store_volinfo (glusterd_volinfo_t *volinfo, >> glusterd_volinfo_ver_ac_t ac) >> { >> int32_t ret = -1; >> >> GF_ASSERT (volinfo) >> >> ret = access("/var/lib/glusterd/vols/gv0/info", F_OK); >> if(ret < 0) >> { >> gf_msg (THIS->name, GF_LOG_ERROR, 0, 0, "info is not >> exit(%d)", errno); >> } >> else >> { >> ret = stat("/var/lib/glusterd/vols/gv0/info", ); >> if(ret < 0) >> { >> gf_msg (THIS->name, GF_LOG_ERROR, 0, 0, "stat >> info error"); >> } >> else >> { >> gf_msg (THIS->name, GF_LOG_ERROR, 0, 0, "info >> size is %lu, inode num is %lu", buf.st_size, buf.st_ino); >> } >> } >> >> glusterd_perform_volinfo_version_action (volinfo, ac); >> ret = glusterd_store_create_volume_dir (volinfo); >> if (ret) >> goto out; >> >> ... >> } >> >> So it is easy to understand why the info or 10.32.1.144.-opt-lvmdir-c2- >> brick sometimes is empty. >> It is becaue the info file is not exist, and it will be create by “fd = >> open (path, O_RDWR | O_CREAT | O_APPEND, 0600);” in function >> gf_store_handle_new. >> And the info file is empty before rename. >> So the info file is empty if glusterd shutdown before rename. >> >> > >> My question is following. >> 1.I did not find the point the info is removed.Could you tell me the >> point where the info and /bricks/* are removed? >> 2.why the file info and bricks/* is removed?But other files in >> var/lib/glusterd/vols/xxx/ >> are not be removed? >> > > AFAIK, we never delete the info file and hence this file is opened with > O_APPEND flag. As I said I will go back and cross check the code once again. > > > > >> Thanks, >> Xin >> >> >> 在 2016-11-11 20:34:05,"Atin Mukherjee" 写道: >> >> >> >> On Fri, Nov 11, 2016 at 4:00 PM, songxin wrote: >> >>> Hi Atin, >>> >>> Thank you for your support. >>> Sincerely wait for your reply. >>> >>> By the way, could you make sure that the issue, file info is empty, >>> cause by rename is interrupted in kernel? >>> >> >> As per my RCA on that bug, it looked to be. >> >> >>> >>> Thanks, >>> Xin >>> >>> 在 2016-11-11 15:49:02,"Atin Mukherjee" 写道: >>> >>> >>> >>> On Fri, Nov 11, 2016 at 1:15 PM, songxin
Re: [Gluster-users] question about info and info.tmp
Hi Atin, I think the root cause is in the function glusterd_import_friend_volume as below. int32_t glusterd_import_friend_volume (dict_t *peer_data, size_t count) { ... ret = glusterd_volinfo_find (new_volinfo->volname, _volinfo); if (0 == ret) { (void) gd_check_and_update_rebalance_info (old_volinfo, new_volinfo); (void) glusterd_delete_stale_volume (old_volinfo, new_volinfo); } ... ret = glusterd_store_volinfo (new_volinfo, GLUSTERD_VOLINFO_VER_AC_NONE); if (ret) { gf_msg (this->name, GF_LOG_ERROR, 0, GD_MSG_VOLINFO_STORE_FAIL, "Failed to store " "volinfo for volume %s", new_volinfo->volname); goto out; } ... } glusterd_delete_stale_volume will remove the info and bricks/* and the glusterd_store_volinfo will create the new one. But if glusterd is killed before rename the info will is empty. And glusterd will start failed because the infois empty in the next time you start the glusterd. Any idea, Atin? Thanks, Xin 在 2016-11-15 12:07:05,"Atin Mukherjee"写道: On Tue, Nov 15, 2016 at 8:58 AM, songxin wrote: Hi Atin, I have some clues about this issue. I could reproduce this issue use the scrip that mentioned in https://bugzilla.redhat.com/show_bug.cgi?id=1308487 . I really appreciate your help in trying to nail down this issue. While I am at your email and going through the code to figure out the possible cause for it, unfortunately I don't see any script in the attachment of the bug. Could you please cross check? After I added some debug print,which like below, in glusterd-store.c and I found that the /var/lib/glusterd/vols/xxx/info and /var/lib/glusterd/vols/xxx/bricks/* are removed. But other files in /var/lib/glusterd/vols/xxx/ will not be remove. int32_t glusterd_store_volinfo (glusterd_volinfo_t *volinfo, glusterd_volinfo_ver_ac_t ac) { int32_t ret = -1; GF_ASSERT (volinfo) ret = access("/var/lib/glusterd/vols/gv0/info", F_OK); if(ret < 0) { gf_msg (THIS->name, GF_LOG_ERROR, 0, 0, "info is not exit(%d)", errno); } else { ret = stat("/var/lib/glusterd/vols/gv0/info", ); if(ret < 0) { gf_msg (THIS->name, GF_LOG_ERROR, 0, 0, "stat info error"); } else { gf_msg (THIS->name, GF_LOG_ERROR, 0, 0, "info size is %lu, inode num is %lu", buf.st_size, buf.st_ino); } } glusterd_perform_volinfo_version_action (volinfo, ac); ret = glusterd_store_create_volume_dir (volinfo); if (ret) goto out; ... } So it is easy to understand why the info or 10.32.1.144.-opt-lvmdir-c2-brick sometimes is empty. It is becaue the info file is not exist, and it will be create by “fd = open (path, O_RDWR | O_CREAT | O_APPEND, 0600);” in function gf_store_handle_new. And the info file is empty before rename. So the info file is empty if glusterd shutdown before rename. My question is following. 1.I did not find the point the info is removed.Could you tell me the point where the info and /bricks/* are removed? 2.why the file info and bricks/* is removed?But other files in var/lib/glusterd/vols/xxx/ are not be removed? AFAIK, we never delete the info file and hence this file is opened with O_APPEND flag. As I said I will go back and cross check the code once again. Thanks, Xin 在 2016-11-11 20:34:05,"Atin Mukherjee" 写道: On Fri, Nov 11, 2016 at 4:00 PM, songxin wrote: Hi Atin, Thank you for your support. Sincerely wait for your reply. By the way, could you make sure that the issue, file info is empty, cause by rename is interrupted in kernel? As per my RCA on that bug, it looked to be. Thanks, Xin 在 2016-11-11 15:49:02,"Atin Mukherjee" 写道: On Fri, Nov 11, 2016 at 1:15 PM, songxin wrote: Hi Atin, Thank you for your reply. Actually it is very difficult to reproduce because I don't know when there was an ongoing commit happening.It is just a coincidence. But I want to make sure the root cause. I'll give it a another try and see if this situation can be simulated/reproduced and will keep you posted. So I would be grateful if you could answer my questions below. You said that "This issue is hit at part of the negative testing where while gluster volume set was executed at the same point of time glusterd in another instance was brought down. In the faulty node we could see /var/lib/glusterd/vols/info file been empty whereas the info.tmp file has the correct
Re: [Gluster-users] question about info and info.tmp
Hi Atin, Now I have known that the info and bricks/* are removed by the function glusterd_delete_stale_volume(). But I have not known how to solve this issue. Thanks, Xin 在 2016-11-15 12:07:05,"Atin Mukherjee"写道: On Tue, Nov 15, 2016 at 8:58 AM, songxin wrote: Hi Atin, I have some clues about this issue. I could reproduce this issue use the scrip that mentioned in https://bugzilla.redhat.com/show_bug.cgi?id=1308487 . I really appreciate your help in trying to nail down this issue. While I am at your email and going through the code to figure out the possible cause for it, unfortunately I don't see any script in the attachment of the bug. Could you please cross check? After I added some debug print,which like below, in glusterd-store.c and I found that the /var/lib/glusterd/vols/xxx/info and /var/lib/glusterd/vols/xxx/bricks/* are removed. But other files in /var/lib/glusterd/vols/xxx/ will not be remove. int32_t glusterd_store_volinfo (glusterd_volinfo_t *volinfo, glusterd_volinfo_ver_ac_t ac) { int32_t ret = -1; GF_ASSERT (volinfo) ret = access("/var/lib/glusterd/vols/gv0/info", F_OK); if(ret < 0) { gf_msg (THIS->name, GF_LOG_ERROR, 0, 0, "info is not exit(%d)", errno); } else { ret = stat("/var/lib/glusterd/vols/gv0/info", ); if(ret < 0) { gf_msg (THIS->name, GF_LOG_ERROR, 0, 0, "stat info error"); } else { gf_msg (THIS->name, GF_LOG_ERROR, 0, 0, "info size is %lu, inode num is %lu", buf.st_size, buf.st_ino); } } glusterd_perform_volinfo_version_action (volinfo, ac); ret = glusterd_store_create_volume_dir (volinfo); if (ret) goto out; ... } So it is easy to understand why the info or 10.32.1.144.-opt-lvmdir-c2-brick sometimes is empty. It is becaue the info file is not exist, and it will be create by “fd = open (path, O_RDWR | O_CREAT | O_APPEND, 0600);” in function gf_store_handle_new. And the info file is empty before rename. So the info file is empty if glusterd shutdown before rename. My question is following. 1.I did not find the point the info is removed.Could you tell me the point where the info and /bricks/* are removed? 2.why the file info and bricks/* is removed?But other files in var/lib/glusterd/vols/xxx/ are not be removed? AFAIK, we never delete the info file and hence this file is opened with O_APPEND flag. As I said I will go back and cross check the code once again. Thanks, Xin 在 2016-11-11 20:34:05,"Atin Mukherjee" 写道: On Fri, Nov 11, 2016 at 4:00 PM, songxin wrote: Hi Atin, Thank you for your support. Sincerely wait for your reply. By the way, could you make sure that the issue, file info is empty, cause by rename is interrupted in kernel? As per my RCA on that bug, it looked to be. Thanks, Xin 在 2016-11-11 15:49:02,"Atin Mukherjee" 写道: On Fri, Nov 11, 2016 at 1:15 PM, songxin wrote: Hi Atin, Thank you for your reply. Actually it is very difficult to reproduce because I don't know when there was an ongoing commit happening.It is just a coincidence. But I want to make sure the root cause. I'll give it a another try and see if this situation can be simulated/reproduced and will keep you posted. So I would be grateful if you could answer my questions below. You said that "This issue is hit at part of the negative testing where while gluster volume set was executed at the same point of time glusterd in another instance was brought down. In the faulty node we could see /var/lib/glusterd/vols/info file been empty whereas the info.tmp file has the correct contents." in comment. I have two questions for you. 1.Could you reproduce this issue by gluster volume set glusterd which was brought down? 2.Could you be certain that this issue is cause by rename is interrupted in kernel? In my case there are two files, info and 10.32.1.144.-opt-lvmdir-c2-brick, are both empty. But in my view only one rename can be running at the same time because of the big lock. Why there are two files are empty? Could rename("info.tmp", "info") and rename("xxx-brick.tmp", "xxx-brick") be running in two thread? Thanks, Xin 在 2016-11-11 15:27:03,"Atin Mukherjee" 写道: On Fri, Nov 11, 2016 at 12:38 PM, songxin wrote: Hi Atin, Thank you for your reply. As you said that the info file can only be changed in the glusterd_store_volinfo() sequentially because of the big lock. I have found the similar issue as below that you mentioned. https://bugzilla.redhat.com/show_bug.cgi?id=1308487 Great, so this is what I was
Re: [Gluster-users] question about info and info.tmp
Hi Atin, I have two nodes, a node and b node, in which creating a replicate volume and then start the volume. I will run the script as below on b node. #!/bin/bash i=1 while(($i<100)) do gluster volume set gv0 nfs.disable on sleep 2s gluster volume set gv0 nfs.disable off i=$(($i+1)) done And I run the script as below on a node at the same time. #!/bin/bash i=1 while(($i<100)) do systemctl stop glusterd systemctl start glusterd gluster volume info i=$(($i+1)) done The issue is very easy reproduced on a board. Could you please tell me where is the point that info file is unlink? Thanks, Xin 在 2016-11-15 12:07:05,"Atin Mukherjee"写道: On Tue, Nov 15, 2016 at 8:58 AM, songxin wrote: Hi Atin, I have some clues about this issue. I could reproduce this issue use the scrip that mentioned in https://bugzilla.redhat.com/show_bug.cgi?id=1308487 . I really appreciate your help in trying to nail down this issue. While I am at your email and going through the code to figure out the possible cause for it, unfortunately I don't see any script in the attachment of the bug. Could you please cross check? After I added some debug print,which like below, in glusterd-store.c and I found that the /var/lib/glusterd/vols/xxx/info and /var/lib/glusterd/vols/xxx/bricks/* are removed. But other files in /var/lib/glusterd/vols/xxx/ will not be remove. int32_t glusterd_store_volinfo (glusterd_volinfo_t *volinfo, glusterd_volinfo_ver_ac_t ac) { int32_t ret = -1; GF_ASSERT (volinfo) ret = access("/var/lib/glusterd/vols/gv0/info", F_OK); if(ret < 0) { gf_msg (THIS->name, GF_LOG_ERROR, 0, 0, "info is not exit(%d)", errno); } else { ret = stat("/var/lib/glusterd/vols/gv0/info", ); if(ret < 0) { gf_msg (THIS->name, GF_LOG_ERROR, 0, 0, "stat info error"); } else { gf_msg (THIS->name, GF_LOG_ERROR, 0, 0, "info size is %lu, inode num is %lu", buf.st_size, buf.st_ino); } } glusterd_perform_volinfo_version_action (volinfo, ac); ret = glusterd_store_create_volume_dir (volinfo); if (ret) goto out; ... } So it is easy to understand why the info or 10.32.1.144.-opt-lvmdir-c2-brick sometimes is empty. It is becaue the info file is not exist, and it will be create by “fd = open (path, O_RDWR | O_CREAT | O_APPEND, 0600);” in function gf_store_handle_new. And the info file is empty before rename. So the info file is empty if glusterd shutdown before rename. My question is following. 1.I did not find the point the info is removed.Could you tell me the point where the info and /bricks/* are removed? 2.why the file info and bricks/* is removed?But other files in var/lib/glusterd/vols/xxx/ are not be removed? AFAIK, we never delete the info file and hence this file is opened with O_APPEND flag. As I said I will go back and cross check the code once again. Thanks, Xin 在 2016-11-11 20:34:05,"Atin Mukherjee" 写道: On Fri, Nov 11, 2016 at 4:00 PM, songxin wrote: Hi Atin, Thank you for your support. Sincerely wait for your reply. By the way, could you make sure that the issue, file info is empty, cause by rename is interrupted in kernel? As per my RCA on that bug, it looked to be. Thanks, Xin 在 2016-11-11 15:49:02,"Atin Mukherjee" 写道: On Fri, Nov 11, 2016 at 1:15 PM, songxin wrote: Hi Atin, Thank you for your reply. Actually it is very difficult to reproduce because I don't know when there was an ongoing commit happening.It is just a coincidence. But I want to make sure the root cause. I'll give it a another try and see if this situation can be simulated/reproduced and will keep you posted. So I would be grateful if you could answer my questions below. You said that "This issue is hit at part of the negative testing where while gluster volume set was executed at the same point of time glusterd in another instance was brought down. In the faulty node we could see /var/lib/glusterd/vols/info file been empty whereas the info.tmp file has the correct contents." in comment. I have two questions for you. 1.Could you reproduce this issue by gluster volume set glusterd which was brought down? 2.Could you be certain that this issue is cause by rename is interrupted in kernel? In my case there are two files, info and 10.32.1.144.-opt-lvmdir-c2-brick, are both empty. But in my view only one rename can
Re: [Gluster-users] question about info and info.tmp
On Tue, Nov 15, 2016 at 8:58 AM, songxinwrote: > Hi Atin, > I have some clues about this issue. > I could reproduce this issue use the scrip that mentioned in > https://bugzilla.redhat.com/show_bug.cgi?id=1308487 . > I really appreciate your help in trying to nail down this issue. While I am at your email and going through the code to figure out the possible cause for it, unfortunately I don't see any script in the attachment of the bug. Could you please cross check? > > After I added some debug print,which like below, in glusterd-store.c and I > found that the /var/lib/glusterd/vols/xxx/info and > /var/lib/glusterd/vols/xxx/bricks/* > are removed. > But other files in /var/lib/glusterd/vols/xxx/ will not be remove. > > int32_t > glusterd_store_volinfo (glusterd_volinfo_t *volinfo, > glusterd_volinfo_ver_ac_t ac) > { > int32_t ret = -1; > > GF_ASSERT (volinfo) > > ret = access("/var/lib/glusterd/vols/gv0/info", F_OK); > if(ret < 0) > { > gf_msg (THIS->name, GF_LOG_ERROR, 0, 0, "info is not > exit(%d)", errno); > } > else > { > ret = stat("/var/lib/glusterd/vols/gv0/info", ); > if(ret < 0) > { > gf_msg (THIS->name, GF_LOG_ERROR, 0, 0, "stat info > error"); > } > else > { > gf_msg (THIS->name, GF_LOG_ERROR, 0, 0, "info size > is %lu, inode num is %lu", buf.st_size, buf.st_ino); > } > } > > glusterd_perform_volinfo_version_action (volinfo, ac); > ret = glusterd_store_create_volume_dir (volinfo); > if (ret) > goto out; > > ... > } > > So it is easy to understand why the info or 10.32.1.144.-opt-lvmdir-c2- > brick sometimes is empty. > It is becaue the info file is not exist, and it will be create by “fd = > open (path, O_RDWR | O_CREAT | O_APPEND, 0600);” in function > gf_store_handle_new. > And the info file is empty before rename. > So the info file is empty if glusterd shutdown before rename. > > > My question is following. > 1.I did not find the point the info is removed.Could you tell me the point > where the info and /bricks/* are removed? > 2.why the file info and bricks/* is removed?But other files in > var/lib/glusterd/vols/xxx/ > are not be removed? > AFAIK, we never delete the info file and hence this file is opened with O_APPEND flag. As I said I will go back and cross check the code once again. > Thanks, > Xin > > > 在 2016-11-11 20:34:05,"Atin Mukherjee" 写道: > > > > On Fri, Nov 11, 2016 at 4:00 PM, songxin wrote: > >> Hi Atin, >> >> Thank you for your support. >> Sincerely wait for your reply. >> >> By the way, could you make sure that the issue, file info is empty, cause >> by rename is interrupted in kernel? >> > > As per my RCA on that bug, it looked to be. > > >> >> Thanks, >> Xin >> >> 在 2016-11-11 15:49:02,"Atin Mukherjee" 写道: >> >> >> >> On Fri, Nov 11, 2016 at 1:15 PM, songxin wrote: >> >>> Hi Atin, >>> Thank you for your reply. >>> Actually it is very difficult to reproduce because I don't know when there >>> was an ongoing commit happening.It is just a coincidence. >>> But I want to make sure the root cause. >>> >> >> I'll give it a another try and see if this situation can be >> simulated/reproduced and will keep you posted. >> >> >>> >>> So I would be grateful if you could answer my questions below. >>> >>> You said that "This issue is hit at part of the negative testing where >>> while gluster volume set was executed at the same point of time glusterd in >>> another instance was brought down. In the faulty node we could see >>> /var/lib/glusterd/vols/info file been empty whereas the >>> info.tmp file has the correct contents." in comment. >>> >>> I have two questions for you. >>> >>> 1.Could you reproduce this issue by gluster volume set glusterd which was >>> brought down? >>> 2.Could you be certain that this issue is cause by rename is interrupted in >>> kernel? >>> >>> In my case there are two files, info and 10.32.1.144.-opt-lvmdir-c2-brick, >>> are both empty. >>> But in my view only one rename can be running at the same time because of >>> the big lock. >>> Why there are two files are empty? >>> >>> >>> Could rename("info.tmp", "info") and rename("xxx-brick.tmp", "xxx-brick") >>> be running in two thread? >>> >>> Thanks, >>> Xin >>> >>> >>> 在 2016-11-11 15:27:03,"Atin Mukherjee" 写道: >>> >>> >>> >>> On Fri, Nov 11, 2016 at 12:38 PM, songxin wrote: >>> Hi Atin, Thank you for your reply. As you said that the info file can only be changed in the glusterd_store_volinfo() sequentially because of the big lock. I have found the similar issue as below that you
Re: [Gluster-users] question about info and info.tmp
Hi Atin, I have some clues about this issue. I could reproduce this issue use the scrip that mentioned in https://bugzilla.redhat.com/show_bug.cgi?id=1308487 . After I added some debug print,which like below, in glusterd-store.c and I found that the /var/lib/glusterd/vols/xxx/info and /var/lib/glusterd/vols/xxx/bricks/* are removed. But other files in /var/lib/glusterd/vols/xxx/ will not be remove. int32_t glusterd_store_volinfo (glusterd_volinfo_t *volinfo, glusterd_volinfo_ver_ac_t ac) { int32_t ret = -1; GF_ASSERT (volinfo) ret = access("/var/lib/glusterd/vols/gv0/info", F_OK); if(ret < 0) { gf_msg (THIS->name, GF_LOG_ERROR, 0, 0, "info is not exit(%d)", errno); } else { ret = stat("/var/lib/glusterd/vols/gv0/info", ); if(ret < 0) { gf_msg (THIS->name, GF_LOG_ERROR, 0, 0, "stat info error"); } else { gf_msg (THIS->name, GF_LOG_ERROR, 0, 0, "info size is %lu, inode num is %lu", buf.st_size, buf.st_ino); } } glusterd_perform_volinfo_version_action (volinfo, ac); ret = glusterd_store_create_volume_dir (volinfo); if (ret) goto out; ... } So it is easy to understand why the info or 10.32.1.144.-opt-lvmdir-c2-brick sometimes is empty. It is becaue the info file is not exist, and it will be create by “fd = open (path, O_RDWR | O_CREAT | O_APPEND, 0600);” in function gf_store_handle_new. And the info file is empty before rename. So the info file is empty if glusterd shutdown before rename. My question is following. 1.I did not find the point the info is removed.Could you tell me the point where the info and /bricks/* are removed? 2.why the file info and bricks/* is removed?But other files in var/lib/glusterd/vols/xxx/ are not be removed? Thanks, Xin 在 2016-11-11 20:34:05,"Atin Mukherjee"写道: On Fri, Nov 11, 2016 at 4:00 PM, songxin wrote: Hi Atin, Thank you for your support. Sincerely wait for your reply. By the way, could you make sure that the issue, file info is empty, cause by rename is interrupted in kernel? As per my RCA on that bug, it looked to be. Thanks, Xin 在 2016-11-11 15:49:02,"Atin Mukherjee" 写道: On Fri, Nov 11, 2016 at 1:15 PM, songxin wrote: Hi Atin, Thank you for your reply. Actually it is very difficult to reproduce because I don't know when there was an ongoing commit happening.It is just a coincidence. But I want to make sure the root cause. I'll give it a another try and see if this situation can be simulated/reproduced and will keep you posted. So I would be grateful if you could answer my questions below. You said that "This issue is hit at part of the negative testing where while gluster volume set was executed at the same point of time glusterd in another instance was brought down. In the faulty node we could see /var/lib/glusterd/vols/info file been empty whereas the info.tmp file has the correct contents." in comment. I have two questions for you. 1.Could you reproduce this issue by gluster volume set glusterd which was brought down? 2.Could you be certain that this issue is cause by rename is interrupted in kernel? In my case there are two files, info and 10.32.1.144.-opt-lvmdir-c2-brick, are both empty. But in my view only one rename can be running at the same time because of the big lock. Why there are two files are empty? Could rename("info.tmp", "info") and rename("xxx-brick.tmp", "xxx-brick") be running in two thread? Thanks, Xin 在 2016-11-11 15:27:03,"Atin Mukherjee" 写道: On Fri, Nov 11, 2016 at 12:38 PM, songxin wrote: Hi Atin, Thank you for your reply. As you said that the info file can only be changed in the glusterd_store_volinfo() sequentially because of the big lock. I have found the similar issue as below that you mentioned. https://bugzilla.redhat.com/show_bug.cgi?id=1308487 Great, so this is what I was actually trying to refer in my first email that I saw a similar issue. Have you got a chance to look at https://bugzilla.redhat.com/show_bug.cgi?id=1308487#c4 ? But in your case, did you try to bring down glusterd when there was an ongoing commit happening? You said that "This issue is hit at part of the negative testing where while gluster volume set was executed at the same point of time glusterd in another instance was brought down. In the faulty node we could see /var/lib/glusterd/vols/info file been empty whereas the info.tmp file has the correct contents." in comment. I have two questions for you. 1.Could you reproduce this issue by gluster volume set glusterd which was brought down? 2.Could you be certain that this
Re: [Gluster-users] question about info and info.tmp
On Fri, Nov 11, 2016 at 4:00 PM, songxinwrote: > Hi Atin, > > Thank you for your support. > Sincerely wait for your reply. > > By the way, could you make sure that the issue, file info is empty, cause > by rename is interrupted in kernel? > As per my RCA on that bug, it looked to be. > > Thanks, > Xin > > 在 2016-11-11 15:49:02,"Atin Mukherjee" 写道: > > > > On Fri, Nov 11, 2016 at 1:15 PM, songxin wrote: > >> Hi Atin, >> Thank you for your reply. >> Actually it is very difficult to reproduce because I don't know when there >> was an ongoing commit happening.It is just a coincidence. >> But I want to make sure the root cause. >> > > I'll give it a another try and see if this situation can be > simulated/reproduced and will keep you posted. > > >> >> So I would be grateful if you could answer my questions below. >> >> You said that "This issue is hit at part of the negative testing where >> while gluster volume set was executed at the same point of time glusterd in >> another instance was brought down. In the faulty node we could see >> /var/lib/glusterd/vols/info file been empty whereas the >> info.tmp file has the correct contents." in comment. >> >> I have two questions for you. >> >> 1.Could you reproduce this issue by gluster volume set glusterd which was >> brought down? >> 2.Could you be certain that this issue is cause by rename is interrupted in >> kernel? >> >> In my case there are two files, info and 10.32.1.144.-opt-lvmdir-c2-brick, >> are both empty. >> But in my view only one rename can be running at the same time because of >> the big lock. >> Why there are two files are empty? >> >> >> Could rename("info.tmp", "info") and rename("xxx-brick.tmp", "xxx-brick") be >> running in two thread? >> >> Thanks, >> Xin >> >> >> 在 2016-11-11 15:27:03,"Atin Mukherjee" 写道: >> >> >> >> On Fri, Nov 11, 2016 at 12:38 PM, songxin wrote: >> >>> >>> Hi Atin, >>> Thank you for your reply. >>> >>> As you said that the info file can only be changed in the >>> glusterd_store_volinfo() >>> sequentially because of the big lock. >>> >>> I have found the similar issue as below that you mentioned. >>> https://bugzilla.redhat.com/show_bug.cgi?id=1308487 >>> >> >> Great, so this is what I was actually trying to refer in my first email >> that I saw a similar issue. Have you got a chance to look at >> https://bugzilla.redhat.com/show_bug.cgi?id=1308487#c4 ? But in your >> case, did you try to bring down glusterd when there was an ongoing commit >> happening? >> >> >>> >>> You said that "This issue is hit at part of the negative testing where >>> while gluster volume set was executed at the same point of time glusterd in >>> another instance was brought down. In the faulty node we could see >>> /var/lib/glusterd/vols/info file been empty whereas the >>> info.tmp file has the correct contents." in comment. >>> >>> I have two questions for you. >>> >>> 1.Could you reproduce this issue by gluster volume set glusterd which was >>> brought down? >>> 2.Could you be certain that this issue is cause by rename is interrupted in >>> kernel? >>> >>> In my case there are two files, info and 10.32.1.144.-opt-lvmdir-c2-brick, >>> are both empty. >>> But in my view only one rename can be running at the same time because of >>> the big lock. >>> Why there are two files are empty? >>> >>> >>> Could rename("info.tmp", "info") and rename("xxx-brick.tmp", "xxx-brick") >>> be running in two thread? >>> >>> Thanks, >>> Xin >>> >>> >>> >>> >>> 在 2016-11-11 14:36:40,"Atin Mukherjee" 写道: >>> >>> >>> >>> On Fri, Nov 11, 2016 at 8:33 AM, songxin wrote: >>> Hi Atin, Thank you for your reply. I have two questions for you. 1.Are the two files info and info.tmp are only to be created or changed in function glusterd_store_volinfo()? I did not find other point in which the two file are changed. >>> >>> If we are talking about info file volume then yes, the mentioned >>> function actually takes care of it. >>> >>> 2.I found that glusterd_store_volinfo() will be call in many point by glusterd.Is there a problem of thread synchronization?If so, one thread may open a same file info.tmp using O_TRUNC flag when another thread is writing the info,tmp.Could this case happen? >>> >>> In glusterd threads are big lock protected and I don't see a >>> possibility (theoretically) to have two glusterd_store_volinfo () calls at >>> a given point of time. >>> >>> Thanks, Xin At 2016-11-10 21:41:06, "Atin Mukherjee" wrote: Did you run out of disk space by any chance? AFAIK, the code is like we write new stuffs to .tmp file and rename it back to the original file. In case of a disk space issue I expect both the files to be of non zero size. But having said that
Re: [Gluster-users] question about info and info.tmp
Hi Atin, Thank you for your support. Sincerely wait for your reply. By the way, could you make sure that the issue, file info is empty, cause by rename is interrupted in kernel? Thanks, Xin 在 2016-11-11 15:49:02,"Atin Mukherjee"写道: On Fri, Nov 11, 2016 at 1:15 PM, songxin wrote: Hi Atin, Thank you for your reply. Actually it is very difficult to reproduce because I don't know when there was an ongoing commit happening.It is just a coincidence. But I want to make sure the root cause. I'll give it a another try and see if this situation can be simulated/reproduced and will keep you posted. So I would be grateful if you could answer my questions below. You said that "This issue is hit at part of the negative testing where while gluster volume set was executed at the same point of time glusterd in another instance was brought down. In the faulty node we could see /var/lib/glusterd/vols/info file been empty whereas the info.tmp file has the correct contents." in comment. I have two questions for you. 1.Could you reproduce this issue by gluster volume set glusterd which was brought down? 2.Could you be certain that this issue is cause by rename is interrupted in kernel? In my case there are two files, info and 10.32.1.144.-opt-lvmdir-c2-brick, are both empty. But in my view only one rename can be running at the same time because of the big lock. Why there are two files are empty? Could rename("info.tmp", "info") and rename("xxx-brick.tmp", "xxx-brick") be running in two thread? Thanks, Xin 在 2016-11-11 15:27:03,"Atin Mukherjee" 写道: On Fri, Nov 11, 2016 at 12:38 PM, songxin wrote: Hi Atin, Thank you for your reply. As you said that the info file can only be changed in the glusterd_store_volinfo() sequentially because of the big lock. I have found the similar issue as below that you mentioned. https://bugzilla.redhat.com/show_bug.cgi?id=1308487 Great, so this is what I was actually trying to refer in my first email that I saw a similar issue. Have you got a chance to look at https://bugzilla.redhat.com/show_bug.cgi?id=1308487#c4 ? But in your case, did you try to bring down glusterd when there was an ongoing commit happening? You said that "This issue is hit at part of the negative testing where while gluster volume set was executed at the same point of time glusterd in another instance was brought down. In the faulty node we could see /var/lib/glusterd/vols/info file been empty whereas the info.tmp file has the correct contents." in comment. I have two questions for you. 1.Could you reproduce this issue by gluster volume set glusterd which was brought down? 2.Could you be certain that this issue is cause by rename is interrupted in kernel? In my case there are two files, info and 10.32.1.144.-opt-lvmdir-c2-brick, are both empty. But in my view only one rename can be running at the same time because of the big lock. Why there are two files are empty? Could rename("info.tmp", "info") and rename("xxx-brick.tmp", "xxx-brick") be running in two thread? Thanks, Xin 在 2016-11-11 14:36:40,"Atin Mukherjee" 写道: On Fri, Nov 11, 2016 at 8:33 AM, songxin wrote: Hi Atin, Thank you for your reply. I have two questions for you. 1.Are the two files info and info.tmp are only to be created or changed in function glusterd_store_volinfo()? I did not find other point in which the two file are changed. If we are talking about info file volume then yes, the mentioned function actually takes care of it. 2.I found that glusterd_store_volinfo() will be call in many point by glusterd.Is there a problem of thread synchronization?If so, one thread may open a same file info.tmp using O_TRUNC flag when another thread is writing the info,tmp.Could this case happen? In glusterd threads are big lock protected and I don't see a possibility (theoretically) to have two glusterd_store_volinfo () calls at a given point of time. Thanks, Xin At 2016-11-10 21:41:06, "Atin Mukherjee" wrote: Did you run out of disk space by any chance? AFAIK, the code is like we write new stuffs to .tmp file and rename it back to the original file. In case of a disk space issue I expect both the files to be of non zero size. But having said that I vaguely remember a similar issue (in the form of a bug or an email) landed up once but we couldn't reproduce it, so something is wrong with the atomic update here is what I guess. I'll be glad if you have a reproducer for the same and then we can dig into it further. On Thu, Nov 10, 2016 at 1:32 PM, songxin wrote: Hi, When I start the glusterd some error happened. And the log is following. [2016-11-08 07:58:34.989365] I [MSGID: 100030] [glusterfsd.c:2318:main] 0-/usr/sbin/glusterd: Started running /usr/sbin/glusterd version
Re: [Gluster-users] question about info and info.tmp
On Fri, Nov 11, 2016 at 1:15 PM, songxinwrote: > Hi Atin, > Thank you for your reply. > Actually it is very difficult to reproduce because I don't know when there > was an ongoing commit happening.It is just a coincidence. > But I want to make sure the root cause. > I'll give it a another try and see if this situation can be simulated/reproduced and will keep you posted. > > So I would be grateful if you could answer my questions below. > > You said that "This issue is hit at part of the negative testing where > while gluster volume set was executed at the same point of time glusterd in > another instance was brought down. In the faulty node we could see > /var/lib/glusterd/vols/info file been empty whereas the info.tmp > file has the correct contents." in comment. > > I have two questions for you. > > 1.Could you reproduce this issue by gluster volume set glusterd which was > brought down? > 2.Could you be certain that this issue is cause by rename is interrupted in > kernel? > > In my case there are two files, info and 10.32.1.144.-opt-lvmdir-c2-brick, > are both empty. > But in my view only one rename can be running at the same time because of the > big lock. > Why there are two files are empty? > > > Could rename("info.tmp", "info") and rename("xxx-brick.tmp", "xxx-brick") be > running in two thread? > > Thanks, > Xin > > > 在 2016-11-11 15:27:03,"Atin Mukherjee" 写道: > > > > On Fri, Nov 11, 2016 at 12:38 PM, songxin wrote: > >> >> Hi Atin, >> Thank you for your reply. >> >> As you said that the info file can only be changed in the >> glusterd_store_volinfo() >> sequentially because of the big lock. >> >> I have found the similar issue as below that you mentioned. >> https://bugzilla.redhat.com/show_bug.cgi?id=1308487 >> > > Great, so this is what I was actually trying to refer in my first email > that I saw a similar issue. Have you got a chance to look at > https://bugzilla.redhat.com/show_bug.cgi?id=1308487#c4 ? But in your > case, did you try to bring down glusterd when there was an ongoing commit > happening? > > >> >> You said that "This issue is hit at part of the negative testing where >> while gluster volume set was executed at the same point of time glusterd in >> another instance was brought down. In the faulty node we could see >> /var/lib/glusterd/vols/info file been empty whereas the >> info.tmp file has the correct contents." in comment. >> >> I have two questions for you. >> >> 1.Could you reproduce this issue by gluster volume set glusterd which was >> brought down? >> 2.Could you be certain that this issue is cause by rename is interrupted in >> kernel? >> >> In my case there are two files, info and 10.32.1.144.-opt-lvmdir-c2-brick, >> are both empty. >> But in my view only one rename can be running at the same time because of >> the big lock. >> Why there are two files are empty? >> >> >> Could rename("info.tmp", "info") and rename("xxx-brick.tmp", "xxx-brick") be >> running in two thread? >> >> Thanks, >> Xin >> >> >> >> >> 在 2016-11-11 14:36:40,"Atin Mukherjee" 写道: >> >> >> >> On Fri, Nov 11, 2016 at 8:33 AM, songxin wrote: >> >>> Hi Atin, >>> >>> Thank you for your reply. >>> I have two questions for you. >>> >>> 1.Are the two files info and info.tmp are only to be created or changed >>> in function glusterd_store_volinfo()? I did not find other point in which >>> the two file are changed. >>> >> >> If we are talking about info file volume then yes, the mentioned function >> actually takes care of it. >> >> >>> 2.I found that glusterd_store_volinfo() will be call in many point by >>> glusterd.Is there a problem of thread synchronization?If so, one thread may >>> open a same file info.tmp using O_TRUNC flag when another thread is >>> writing the info,tmp.Could this case happen? >>> >> >> In glusterd threads are big lock protected and I don't see a possibility >> (theoretically) to have two glusterd_store_volinfo () calls at a given >> point of time. >> >> >>> >>> Thanks, >>> Xin >>> >>> >>> At 2016-11-10 21:41:06, "Atin Mukherjee" wrote: >>> >>> Did you run out of disk space by any chance? AFAIK, the code is like we >>> write new stuffs to .tmp file and rename it back to the original file. In >>> case of a disk space issue I expect both the files to be of non zero size. >>> But having said that I vaguely remember a similar issue (in the form of a >>> bug or an email) landed up once but we couldn't reproduce it, so something >>> is wrong with the atomic update here is what I guess. I'll be glad if you >>> have a reproducer for the same and then we can dig into it further. >>> >>> On Thu, Nov 10, 2016 at 1:32 PM, songxin wrote: >>> Hi, When I start the glusterd some error happened. And the log is following. [2016-11-08 07:58:34.989365] I [MSGID: 100030] [glusterfsd.c:2318:main]
Re: [Gluster-users] question about info and info.tmp
Hi Atin, Thank you for your reply. Actually it is very difficult to reproduce because I don't know when there was an ongoing commit happening.It is just a coincidence. But I want to make sure the root cause. So I would be grateful if you could answer my questions below. You said that "This issue is hit at part of the negative testing where while gluster volume set was executed at the same point of time glusterd in another instance was brought down. In the faulty node we could see /var/lib/glusterd/vols/info file been empty whereas the info.tmp file has the correct contents." in comment. I have two questions for you. 1.Could you reproduce this issue by gluster volume set glusterd which was brought down? 2.Could you be certain that this issue is cause by rename is interrupted in kernel? In my case there are two files, info and 10.32.1.144.-opt-lvmdir-c2-brick, are both empty. But in my view only one rename can be running at the same time because of the big lock. Why there are two files are empty? Could rename("info.tmp", "info") and rename("xxx-brick.tmp", "xxx-brick") be running in two thread? Thanks, Xin 在 2016-11-11 15:27:03,"Atin Mukherjee"写道: On Fri, Nov 11, 2016 at 12:38 PM, songxin wrote: Hi Atin, Thank you for your reply. As you said that the info file can only be changed in the glusterd_store_volinfo() sequentially because of the big lock. I have found the similar issue as below that you mentioned. https://bugzilla.redhat.com/show_bug.cgi?id=1308487 Great, so this is what I was actually trying to refer in my first email that I saw a similar issue. Have you got a chance to look at https://bugzilla.redhat.com/show_bug.cgi?id=1308487#c4 ? But in your case, did you try to bring down glusterd when there was an ongoing commit happening? You said that "This issue is hit at part of the negative testing where while gluster volume set was executed at the same point of time glusterd in another instance was brought down. In the faulty node we could see /var/lib/glusterd/vols/info file been empty whereas the info.tmp file has the correct contents." in comment. I have two questions for you. 1.Could you reproduce this issue by gluster volume set glusterd which was brought down? 2.Could you be certain that this issue is cause by rename is interrupted in kernel? In my case there are two files, info and 10.32.1.144.-opt-lvmdir-c2-brick, are both empty. But in my view only one rename can be running at the same time because of the big lock. Why there are two files are empty? Could rename("info.tmp", "info") and rename("xxx-brick.tmp", "xxx-brick") be running in two thread? Thanks, Xin 在 2016-11-11 14:36:40,"Atin Mukherjee" 写道: On Fri, Nov 11, 2016 at 8:33 AM, songxin wrote: Hi Atin, Thank you for your reply. I have two questions for you. 1.Are the two files info and info.tmp are only to be created or changed in function glusterd_store_volinfo()? I did not find other point in which the two file are changed. If we are talking about info file volume then yes, the mentioned function actually takes care of it. 2.I found that glusterd_store_volinfo() will be call in many point by glusterd.Is there a problem of thread synchronization?If so, one thread may open a same file info.tmp using O_TRUNC flag when another thread is writing the info,tmp.Could this case happen? In glusterd threads are big lock protected and I don't see a possibility (theoretically) to have two glusterd_store_volinfo () calls at a given point of time. Thanks, Xin At 2016-11-10 21:41:06, "Atin Mukherjee" wrote: Did you run out of disk space by any chance? AFAIK, the code is like we write new stuffs to .tmp file and rename it back to the original file. In case of a disk space issue I expect both the files to be of non zero size. But having said that I vaguely remember a similar issue (in the form of a bug or an email) landed up once but we couldn't reproduce it, so something is wrong with the atomic update here is what I guess. I'll be glad if you have a reproducer for the same and then we can dig into it further. On Thu, Nov 10, 2016 at 1:32 PM, songxin wrote: Hi, When I start the glusterd some error happened. And the log is following. [2016-11-08 07:58:34.989365] I [MSGID: 100030] [glusterfsd.c:2318:main] 0-/usr/sbin/glusterd: Started running /usr/sbin/glusterd version 3.7.6 (args: /usr/sbin/glusterd -p /var/run/glusterd.pid --log-level INFO) [2016-11-08 07:58:34.998356] I [MSGID: 106478] [glusterd.c:1350:init] 0-management: Maximum allowed open file descriptors set to 65536 [2016-11-08 07:58:35.000667] I [MSGID: 106479] [glusterd.c:1399:init] 0-management: Using /system/glusterd as working directory [2016-11-08 07:58:35.024508] I [MSGID: 106514] [glusterd-store.c:2075:glusterd_restore_op_version]
Re: [Gluster-users] question about info and info.tmp
On Fri, Nov 11, 2016 at 12:38 PM, songxinwrote: > > Hi Atin, > Thank you for your reply. > > As you said that the info file can only be changed in the > glusterd_store_volinfo() > sequentially because of the big lock. > > I have found the similar issue as below that you mentioned. > https://bugzilla.redhat.com/show_bug.cgi?id=1308487 > Great, so this is what I was actually trying to refer in my first email that I saw a similar issue. Have you got a chance to look at https://bugzilla.redhat.com/show_bug.cgi?id=1308487#c4 ? But in your case, did you try to bring down glusterd when there was an ongoing commit happening? > > You said that "This issue is hit at part of the negative testing where > while gluster volume set was executed at the same point of time glusterd in > another instance was brought down. In the faulty node we could see > /var/lib/glusterd/vols/info file been empty whereas the info.tmp > file has the correct contents." in comment. > > I have two questions for you. > > 1.Could you reproduce this issue by gluster volume set glusterd which was > brought down? > 2.Could you be certain that this issue is cause by rename is interrupted in > kernel? > > In my case there are two files, info and 10.32.1.144.-opt-lvmdir-c2-brick, > are both empty. > But in my view only one rename can be running at the same time because of the > big lock. > Why there are two files are empty? > > > Could rename("info.tmp", "info") and rename("xxx-brick.tmp", "xxx-brick") be > running in two thread? > > Thanks, > Xin > > > > > 在 2016-11-11 14:36:40,"Atin Mukherjee" 写道: > > > > On Fri, Nov 11, 2016 at 8:33 AM, songxin wrote: > >> Hi Atin, >> >> Thank you for your reply. >> I have two questions for you. >> >> 1.Are the two files info and info.tmp are only to be created or changed >> in function glusterd_store_volinfo()? I did not find other point in which >> the two file are changed. >> > > If we are talking about info file volume then yes, the mentioned function > actually takes care of it. > > >> 2.I found that glusterd_store_volinfo() will be call in many point by >> glusterd.Is there a problem of thread synchronization?If so, one thread may >> open a same file info.tmp using O_TRUNC flag when another thread is >> writing the info,tmp.Could this case happen? >> > > In glusterd threads are big lock protected and I don't see a possibility > (theoretically) to have two glusterd_store_volinfo () calls at a given > point of time. > > >> >> Thanks, >> Xin >> >> >> At 2016-11-10 21:41:06, "Atin Mukherjee" wrote: >> >> Did you run out of disk space by any chance? AFAIK, the code is like we >> write new stuffs to .tmp file and rename it back to the original file. In >> case of a disk space issue I expect both the files to be of non zero size. >> But having said that I vaguely remember a similar issue (in the form of a >> bug or an email) landed up once but we couldn't reproduce it, so something >> is wrong with the atomic update here is what I guess. I'll be glad if you >> have a reproducer for the same and then we can dig into it further. >> >> On Thu, Nov 10, 2016 at 1:32 PM, songxin wrote: >> >>> Hi, >>> When I start the glusterd some error happened. >>> And the log is following. >>> >>> [2016-11-08 07:58:34.989365] I [MSGID: 100030] [glusterfsd.c:2318:main] >>> 0-/usr/sbin/glusterd: Started running /usr/sbin/glusterd version 3.7.6 >>> (args: /usr/sbin/glusterd -p /var/run/glusterd.pid --log-level INFO) >>> [2016-11-08 07:58:34.998356] I [MSGID: 106478] [glusterd.c:1350:init] >>> 0-management: Maximum allowed open file descriptors set to 65536 >>> [2016-11-08 07:58:35.000667] I [MSGID: 106479] [glusterd.c:1399:init] >>> 0-management: Using /system/glusterd as working directory >>> [2016-11-08 07:58:35.024508] I [MSGID: 106514] >>> [glusterd-store.c:2075:glusterd_restore_op_version] 0-management: >>> Upgrade detected. Setting op-version to minimum : 1 >>> *[2016-11-08 07:58:35.025356] E [MSGID: 106206] >>> [glusterd-store.c:2562:glusterd_store_update_volinfo] 0-management: Failed >>> to get next store iter * >>> *[2016-11-08 07:58:35.025401] E [MSGID: 106207] >>> [glusterd-store.c:2844:glusterd_store_retrieve_volume] 0-management: Failed >>> to update volinfo for c_glusterfs volume * >>> *[2016-11-08 07:58:35.025463] E [MSGID: 106201] >>> [glusterd-store.c:3042:glusterd_store_retrieve_volumes] 0-management: >>> Unable to restore volume: c_glusterfs * >>> *[2016-11-08 07:58:35.025544] E [MSGID: 101019] >>> [xlator.c:428:xlator_init] 0-management: Initialization of volume >>> 'management' failed, review your volfile again * >>> *[2016-11-08 07:58:35.025582] E [graph.c:322:glusterfs_graph_init] >>> 0-management: initializing translator failed * >>> *[2016-11-08 07:58:35.025629] E [graph.c:661:glusterfs_graph_activate] >>> 0-graph: init failed * >>> [2016-11-08 07:58:35.026109] W
Re: [Gluster-users] question about info and info.tmp
Hi Atin, Thank you for your reply. As you said that the info file can only be changed in the glusterd_store_volinfo() sequentially because of the big lock. I have found the similar issue as below that you mentioned. https://bugzilla.redhat.com/show_bug.cgi?id=1308487 You said that "This issue is hit at part of the negative testing where while gluster volume set was executed at the same point of time glusterd in another instance was brought down. In the faulty node we could see /var/lib/glusterd/vols/info file been empty whereas the info.tmp file has the correct contents." in comment. I have two questions for you. 1.Could you reproduce this issue by gluster volume set glusterd which was brought down? 2.Could you be certain that this issue is cause by rename is interrupted in kernel? In my case there are two files, info and 10.32.1.144.-opt-lvmdir-c2-brick, are both empty. But in my view only one rename can be running at the same time because of the big lock. Why there are two files are empty? Could rename("info.tmp", "info") and rename("xxx-brick.tmp", "xxx-brick") be running in two thread? Thanks, Xin 在 2016-11-11 14:36:40,"Atin Mukherjee"写道: On Fri, Nov 11, 2016 at 8:33 AM, songxin wrote: Hi Atin, Thank you for your reply. I have two questions for you. 1.Are the two files info and info.tmp are only to be created or changed in function glusterd_store_volinfo()? I did not find other point in which the two file are changed. If we are talking about info file volume then yes, the mentioned function actually takes care of it. 2.I found that glusterd_store_volinfo() will be call in many point by glusterd.Is there a problem of thread synchronization?If so, one thread may open a same file info.tmp using O_TRUNC flag when another thread is writing the info,tmp.Could this case happen? In glusterd threads are big lock protected and I don't see a possibility (theoretically) to have two glusterd_store_volinfo () calls at a given point of time. Thanks, Xin At 2016-11-10 21:41:06, "Atin Mukherjee" wrote: Did you run out of disk space by any chance? AFAIK, the code is like we write new stuffs to .tmp file and rename it back to the original file. In case of a disk space issue I expect both the files to be of non zero size. But having said that I vaguely remember a similar issue (in the form of a bug or an email) landed up once but we couldn't reproduce it, so something is wrong with the atomic update here is what I guess. I'll be glad if you have a reproducer for the same and then we can dig into it further. On Thu, Nov 10, 2016 at 1:32 PM, songxin wrote: Hi, When I start the glusterd some error happened. And the log is following. [2016-11-08 07:58:34.989365] I [MSGID: 100030] [glusterfsd.c:2318:main] 0-/usr/sbin/glusterd: Started running /usr/sbin/glusterd version 3.7.6 (args: /usr/sbin/glusterd -p /var/run/glusterd.pid --log-level INFO) [2016-11-08 07:58:34.998356] I [MSGID: 106478] [glusterd.c:1350:init] 0-management: Maximum allowed open file descriptors set to 65536 [2016-11-08 07:58:35.000667] I [MSGID: 106479] [glusterd.c:1399:init] 0-management: Using /system/glusterd as working directory [2016-11-08 07:58:35.024508] I [MSGID: 106514] [glusterd-store.c:2075:glusterd_restore_op_version] 0-management: Upgrade detected. Setting op-version to minimum : 1 [2016-11-08 07:58:35.025356] E [MSGID: 106206] [glusterd-store.c:2562:glusterd_store_update_volinfo] 0-management: Failed to get next store iter [2016-11-08 07:58:35.025401] E [MSGID: 106207] [glusterd-store.c:2844:glusterd_store_retrieve_volume] 0-management: Failed to update volinfo for c_glusterfs volume [2016-11-08 07:58:35.025463] E [MSGID: 106201] [glusterd-store.c:3042:glusterd_store_retrieve_volumes] 0-management: Unable to restore volume: c_glusterfs [2016-11-08 07:58:35.025544] E [MSGID: 101019] [xlator.c:428:xlator_init] 0-management: Initialization of volume 'management' failed, review your volfile again [2016-11-08 07:58:35.025582] E [graph.c:322:glusterfs_graph_init] 0-management: initializing translator failed [2016-11-08 07:58:35.025629] E [graph.c:661:glusterfs_graph_activate] 0-graph: init failed [2016-11-08 07:58:35.026109] W [glusterfsd.c:1236:cleanup_and_exit] (-->/usr/sbin/glusterd(glusterfs_volumes_init-0x1b260) [0x1000a718] -->/usr/sbin/glusterd(glusterfs_process_volfp-0x1b3b8) [0x1000a5a8] -->/usr/sbin/glusterd(cleanup_and_exit-0x1c02c) [0x100098bc] ) 0-: received signum (0), shutting down And then I found that the size of vols/volume_name/info is 0.It cause glusterd shutdown. But I found that vols/volume_name_info.tmp is not 0. And I found that there is a brick file vols/volume_name/bricks/.brick is 0, but vols/volume_name/bricks/.brick.tmp is not 0. I read the function code glusterd_store_volinfo () in glusterd-store.c . I know that
Re: [Gluster-users] question about info and info.tmp
On Fri, Nov 11, 2016 at 8:33 AM, songxinwrote: > Hi Atin, > > Thank you for your reply. > I have two questions for you. > > 1.Are the two files info and info.tmp are only to be created or changed in > function glusterd_store_volinfo()? I did not find other point in which the > two file are changed. > If we are talking about info file volume then yes, the mentioned function actually takes care of it. > 2.I found that glusterd_store_volinfo() will be call in many point by > glusterd.Is there a problem of thread synchronization?If so, one thread may > open a same file info.tmp using O_TRUNC flag when another thread is > writing the info,tmp.Could this case happen? > In glusterd threads are big lock protected and I don't see a possibility (theoretically) to have two glusterd_store_volinfo () calls at a given point of time. > > Thanks, > Xin > > > At 2016-11-10 21:41:06, "Atin Mukherjee" wrote: > > Did you run out of disk space by any chance? AFAIK, the code is like we > write new stuffs to .tmp file and rename it back to the original file. In > case of a disk space issue I expect both the files to be of non zero size. > But having said that I vaguely remember a similar issue (in the form of a > bug or an email) landed up once but we couldn't reproduce it, so something > is wrong with the atomic update here is what I guess. I'll be glad if you > have a reproducer for the same and then we can dig into it further. > > On Thu, Nov 10, 2016 at 1:32 PM, songxin wrote: > >> Hi, >> When I start the glusterd some error happened. >> And the log is following. >> >> [2016-11-08 07:58:34.989365] I [MSGID: 100030] [glusterfsd.c:2318:main] >> 0-/usr/sbin/glusterd: Started running /usr/sbin/glusterd version 3.7.6 >> (args: /usr/sbin/glusterd -p /var/run/glusterd.pid --log-level INFO) >> [2016-11-08 07:58:34.998356] I [MSGID: 106478] [glusterd.c:1350:init] >> 0-management: Maximum allowed open file descriptors set to 65536 >> [2016-11-08 07:58:35.000667] I [MSGID: 106479] [glusterd.c:1399:init] >> 0-management: Using /system/glusterd as working directory >> [2016-11-08 07:58:35.024508] I [MSGID: 106514] >> [glusterd-store.c:2075:glusterd_restore_op_version] 0-management: >> Upgrade detected. Setting op-version to minimum : 1 >> *[2016-11-08 07:58:35.025356] E [MSGID: 106206] >> [glusterd-store.c:2562:glusterd_store_update_volinfo] 0-management: Failed >> to get next store iter * >> *[2016-11-08 07:58:35.025401] E [MSGID: 106207] >> [glusterd-store.c:2844:glusterd_store_retrieve_volume] 0-management: Failed >> to update volinfo for c_glusterfs volume * >> *[2016-11-08 07:58:35.025463] E [MSGID: 106201] >> [glusterd-store.c:3042:glusterd_store_retrieve_volumes] 0-management: >> Unable to restore volume: c_glusterfs * >> *[2016-11-08 07:58:35.025544] E [MSGID: 101019] >> [xlator.c:428:xlator_init] 0-management: Initialization of volume >> 'management' failed, review your volfile again * >> *[2016-11-08 07:58:35.025582] E [graph.c:322:glusterfs_graph_init] >> 0-management: initializing translator failed * >> *[2016-11-08 07:58:35.025629] E [graph.c:661:glusterfs_graph_activate] >> 0-graph: init failed * >> [2016-11-08 07:58:35.026109] W [glusterfsd.c:1236:cleanup_and_exit] >> (-->/usr/sbin/glusterd(glusterfs_volumes_init-0x1b260) [0x1000a718] >> -->/usr/sbin/glusterd(glusterfs_process_volfp-0x1b3b8) [0x1000a5a8] >> -->/usr/sbin/glusterd(cleanup_and_exit-0x1c02c) [0x100098bc] ) 0-: >> received signum (0), shutting down >> >> >> And then I found that the size of vols/volume_name/info is 0.It cause >> glusterd shutdown. >> But I found that vols/volume_name_info.tmp is not 0. >> And I found that there is a brick file vols/volume_name/bricks/.brick >> is 0, but vols/volume_name/bricks/.brick.tmp is not 0. >> >> I read the function code glusterd_store_volinfo () in glusterd-store.c . >> I know that the info.tmp will be rename to info in function >> glusterd_store_volume_atomic_update(). >> >> But my question is that why the info file is 0 but info.tmp is not 0. >> >> >> Thanks, >> Xin >> >> >> >> >> ___ >> Gluster-users mailing list >> Gluster-users@gluster.org >> http://www.gluster.org/mailman/listinfo/gluster-users >> > > > > -- > > ~ Atin (atinm) > > > > > -- ~ Atin (atinm) ___ Gluster-users mailing list Gluster-users@gluster.org http://www.gluster.org/mailman/listinfo/gluster-users
Re: [Gluster-users] question about info and info.tmp
Hi Atin, Thank you for your reply. I have two questions for you. 1.Are the two files info and info.tmp are only to be created or changed in function glusterd_store_volinfo()? I did not find other point in which the two file are changed. 2.I found that glusterd_store_volinfo() will be call in many point by glusterd.Is there a problem of thread synchronization?If so, one thread may open a same file info.tmp using O_TRUNC flag when another thread is writing the info,tmp.Could this case happen? Thanks, Xin At 2016-11-10 21:41:06, "Atin Mukherjee"wrote: Did you run out of disk space by any chance? AFAIK, the code is like we write new stuffs to .tmp file and rename it back to the original file. In case of a disk space issue I expect both the files to be of non zero size. But having said that I vaguely remember a similar issue (in the form of a bug or an email) landed up once but we couldn't reproduce it, so something is wrong with the atomic update here is what I guess. I'll be glad if you have a reproducer for the same and then we can dig into it further. On Thu, Nov 10, 2016 at 1:32 PM, songxin wrote: Hi, When I start the glusterd some error happened. And the log is following. [2016-11-08 07:58:34.989365] I [MSGID: 100030] [glusterfsd.c:2318:main] 0-/usr/sbin/glusterd: Started running /usr/sbin/glusterd version 3.7.6 (args: /usr/sbin/glusterd -p /var/run/glusterd.pid --log-level INFO) [2016-11-08 07:58:34.998356] I [MSGID: 106478] [glusterd.c:1350:init] 0-management: Maximum allowed open file descriptors set to 65536 [2016-11-08 07:58:35.000667] I [MSGID: 106479] [glusterd.c:1399:init] 0-management: Using /system/glusterd as working directory [2016-11-08 07:58:35.024508] I [MSGID: 106514] [glusterd-store.c:2075:glusterd_restore_op_version] 0-management: Upgrade detected. Setting op-version to minimum : 1 [2016-11-08 07:58:35.025356] E [MSGID: 106206] [glusterd-store.c:2562:glusterd_store_update_volinfo] 0-management: Failed to get next store iter [2016-11-08 07:58:35.025401] E [MSGID: 106207] [glusterd-store.c:2844:glusterd_store_retrieve_volume] 0-management: Failed to update volinfo for c_glusterfs volume [2016-11-08 07:58:35.025463] E [MSGID: 106201] [glusterd-store.c:3042:glusterd_store_retrieve_volumes] 0-management: Unable to restore volume: c_glusterfs [2016-11-08 07:58:35.025544] E [MSGID: 101019] [xlator.c:428:xlator_init] 0-management: Initialization of volume 'management' failed, review your volfile again [2016-11-08 07:58:35.025582] E [graph.c:322:glusterfs_graph_init] 0-management: initializing translator failed [2016-11-08 07:58:35.025629] E [graph.c:661:glusterfs_graph_activate] 0-graph: init failed [2016-11-08 07:58:35.026109] W [glusterfsd.c:1236:cleanup_and_exit] (-->/usr/sbin/glusterd(glusterfs_volumes_init-0x1b260) [0x1000a718] -->/usr/sbin/glusterd(glusterfs_process_volfp-0x1b3b8) [0x1000a5a8] -->/usr/sbin/glusterd(cleanup_and_exit-0x1c02c) [0x100098bc] ) 0-: received signum (0), shutting down And then I found that the size of vols/volume_name/info is 0.It cause glusterd shutdown. But I found that vols/volume_name_info.tmp is not 0. And I found that there is a brick file vols/volume_name/bricks/.brick is 0, but vols/volume_name/bricks/.brick.tmp is not 0. I read the function code glusterd_store_volinfo () in glusterd-store.c . I know that the info.tmp will be rename to info in function glusterd_store_volume_atomic_update(). But my question is that why the info file is 0 but info.tmp is not 0. Thanks, Xin ___ Gluster-users mailing list Gluster-users@gluster.org http://www.gluster.org/mailman/listinfo/gluster-users -- ~ Atin (atinm) ___ Gluster-users mailing list Gluster-users@gluster.org http://www.gluster.org/mailman/listinfo/gluster-users
Re: [Gluster-users] question about info and info.tmp
Did you run out of disk space by any chance? AFAIK, the code is like we write new stuffs to .tmp file and rename it back to the original file. In case of a disk space issue I expect both the files to be of non zero size. But having said that I vaguely remember a similar issue (in the form of a bug or an email) landed up once but we couldn't reproduce it, so something is wrong with the atomic update here is what I guess. I'll be glad if you have a reproducer for the same and then we can dig into it further. On Thu, Nov 10, 2016 at 1:32 PM, songxinwrote: > Hi, > When I start the glusterd some error happened. > And the log is following. > > [2016-11-08 07:58:34.989365] I [MSGID: 100030] [glusterfsd.c:2318:main] > 0-/usr/sbin/glusterd: Started running /usr/sbin/glusterd version 3.7.6 > (args: /usr/sbin/glusterd -p /var/run/glusterd.pid --log-level INFO) > [2016-11-08 07:58:34.998356] I [MSGID: 106478] [glusterd.c:1350:init] > 0-management: Maximum allowed open file descriptors set to 65536 > [2016-11-08 07:58:35.000667] I [MSGID: 106479] [glusterd.c:1399:init] > 0-management: Using /system/glusterd as working directory > [2016-11-08 07:58:35.024508] I [MSGID: 106514] > [glusterd-store.c:2075:glusterd_restore_op_version] > 0-management: Upgrade detected. Setting op-version to minimum : 1 > *[2016-11-08 07:58:35.025356] E [MSGID: 106206] > [glusterd-store.c:2562:glusterd_store_update_volinfo] 0-management: Failed > to get next store iter * > *[2016-11-08 07:58:35.025401] E [MSGID: 106207] > [glusterd-store.c:2844:glusterd_store_retrieve_volume] 0-management: Failed > to update volinfo for c_glusterfs volume * > *[2016-11-08 07:58:35.025463] E [MSGID: 106201] > [glusterd-store.c:3042:glusterd_store_retrieve_volumes] 0-management: > Unable to restore volume: c_glusterfs * > *[2016-11-08 07:58:35.025544] E [MSGID: 101019] [xlator.c:428:xlator_init] > 0-management: Initialization of volume 'management' failed, review your > volfile again * > *[2016-11-08 07:58:35.025582] E [graph.c:322:glusterfs_graph_init] > 0-management: initializing translator failed * > *[2016-11-08 07:58:35.025629] E [graph.c:661:glusterfs_graph_activate] > 0-graph: init failed * > [2016-11-08 07:58:35.026109] W [glusterfsd.c:1236:cleanup_and_exit] > (-->/usr/sbin/glusterd(glusterfs_volumes_init-0x1b260) [0x1000a718] > -->/usr/sbin/glusterd(glusterfs_process_volfp-0x1b3b8) [0x1000a5a8] > -->/usr/sbin/glusterd(cleanup_and_exit-0x1c02c) [0x100098bc] ) 0-: > received signum (0), shutting down > > > And then I found that the size of vols/volume_name/info is 0.It cause > glusterd shutdown. > But I found that vols/volume_name_info.tmp is not 0. > And I found that there is a brick file vols/volume_name/bricks/.brick > is 0, but vols/volume_name/bricks/.brick.tmp is not 0. > > I read the function code glusterd_store_volinfo () in glusterd-store.c . > I know that the info.tmp will be rename to info in function > glusterd_store_volume_atomic_update(). > > But my question is that why the info file is 0 but info.tmp is not 0. > > > Thanks, > Xin > > > > > ___ > Gluster-users mailing list > Gluster-users@gluster.org > http://www.gluster.org/mailman/listinfo/gluster-users > -- ~ Atin (atinm) ___ Gluster-users mailing list Gluster-users@gluster.org http://www.gluster.org/mailman/listinfo/gluster-users
[Gluster-users] question about info and info.tmp
Hi, When I start the glusterd some error happened. And the log is following. [2016-11-08 07:58:34.989365] I [MSGID: 100030] [glusterfsd.c:2318:main] 0-/usr/sbin/glusterd: Started running /usr/sbin/glusterd version 3.7.6 (args: /usr/sbin/glusterd -p /var/run/glusterd.pid --log-level INFO) [2016-11-08 07:58:34.998356] I [MSGID: 106478] [glusterd.c:1350:init] 0-management: Maximum allowed open file descriptors set to 65536 [2016-11-08 07:58:35.000667] I [MSGID: 106479] [glusterd.c:1399:init] 0-management: Using /system/glusterd as working directory [2016-11-08 07:58:35.024508] I [MSGID: 106514] [glusterd-store.c:2075:glusterd_restore_op_version] 0-management: Upgrade detected. Setting op-version to minimum : 1 [2016-11-08 07:58:35.025356] E [MSGID: 106206] [glusterd-store.c:2562:glusterd_store_update_volinfo] 0-management: Failed to get next store iter [2016-11-08 07:58:35.025401] E [MSGID: 106207] [glusterd-store.c:2844:glusterd_store_retrieve_volume] 0-management: Failed to update volinfo for c_glusterfs volume [2016-11-08 07:58:35.025463] E [MSGID: 106201] [glusterd-store.c:3042:glusterd_store_retrieve_volumes] 0-management: Unable to restore volume: c_glusterfs [2016-11-08 07:58:35.025544] E [MSGID: 101019] [xlator.c:428:xlator_init] 0-management: Initialization of volume 'management' failed, review your volfile again [2016-11-08 07:58:35.025582] E [graph.c:322:glusterfs_graph_init] 0-management: initializing translator failed [2016-11-08 07:58:35.025629] E [graph.c:661:glusterfs_graph_activate] 0-graph: init failed [2016-11-08 07:58:35.026109] W [glusterfsd.c:1236:cleanup_and_exit] (-->/usr/sbin/glusterd(glusterfs_volumes_init-0x1b260) [0x1000a718] -->/usr/sbin/glusterd(glusterfs_process_volfp-0x1b3b8) [0x1000a5a8] -->/usr/sbin/glusterd(cleanup_and_exit-0x1c02c) [0x100098bc] ) 0-: received signum (0), shutting down And then I found that the size of vols/volume_name/info is 0.It cause glusterd shutdown. But I found that vols/volume_name_info.tmp is not 0. And I found that there is a brick file vols/volume_name/bricks/.brick is 0, but vols/volume_name/bricks/.brick.tmp is not 0. I read the function code glusterd_store_volinfo () in glusterd-store.c . I know that the info.tmp will be rename to info in function glusterd_store_volume_atomic_update(). But my question is that why the info file is 0 but info.tmp is not 0. Thanks, Xin___ Gluster-users mailing list Gluster-users@gluster.org http://www.gluster.org/mailman/listinfo/gluster-users